1. Imports et config¶
In [1]:
import tensorflow as tf
from tensorflow.keras.applications import EfficientNetB0
from tensorflow.keras.layers import *
from tensorflow.keras.models import Model
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import cv2
import os
import random
from PIL import Image
import re
import warnings
warnings.filterwarnings('ignore')
# Dark theme
plt.style.use('dark_background')
sns.set_palette("husl")
# Seed pour reproductibilité
SEED = 42
random.seed(SEED)
np.random.seed(SEED)
tf.random.set_seed(SEED)
# HYPERPARAMETRES - Recipe Image Retrieval avec EfficientNet
CONFIG = {
# === IMAGE PARAMETERS ===
'IMG_SIZE': 224, # Optimal pour EfficientNet
'BATCH_SIZE': 32, # Bon équilibre mémoire/performance
# === EMBEDDING PARAMETERS ===
'EMBEDDING_DIM_NATIVE': 1280, # EfficientNet features natives
'EMBEDDING_DIM_CUSTOM': 512, # Custom head output (si fine-tuning)
# === PHASE 1: FEATURE EXTRACTION ===
'USE_NATIVE_FEATURES': True, # True = 1280-dim, False = 512-dim custom
# === PHASE 2: TRANSFER LEARNING ===
'TRANSFER_EPOCHS': 10, # Entraînement head seulement
'TRANSFER_LR': 0.001, # Learning rate plus élevé pour nouvelles couches
'TRANSFER_BATCH_SIZE': 16, # Plus petit pour stabilité
# === PHASE 3: FINE-TUNING ===
'FINETUNE_EPOCHS': 15, # Fine-tuning complet
'FINETUNE_LR': 0.0001, # Learning rate faible pour pré-entraîné
'FINETUNE_UNFREEZE_LAYERS': 20, # Nombre de couches à dégeler
# === TRAINING PARAMETERS ===
'TRIPLET_MARGIN': 0.3, # Augmenté pour meilleure séparation
'DROPOUT_RATE': 0.3, # Régularisation
'WEIGHT_DECAY': 0.0001, # L2 regularization
# === OPTIMIZATION ===
'OPTIMIZER': 'adam', # Standard
'PATIENCE': 5, # Early stopping
'REDUCE_LR_PATIENCE': 3, # Réduction LR
'REDUCE_LR_FACTOR': 0.5, # Facteur réduction
# === DATA AUGMENTATION ===
'USE_AUGMENTATION': True, # Pour fine-tuning
'ROTATION_RANGE': 15, # Rotation aléatoire
'ZOOM_RANGE': 0.1, # Zoom aléatoire
'HORIZONTAL_FLIP': True, # Flip horizontal
# === EVALUATION ===
'SIMILARITY_THRESHOLD': 0.7, # Seuil similarité
'TOP_K_RESULTS': 3, # Nombre résultats retournés
'VALIDATION_SPLIT': 0.2, # Split pour validation
}
# CONFIGS SPÉCIFIQUES PAR PHASE
CONFIG_PHASE1 = {**CONFIG, **{
'MODE': 'feature_extraction',
'EMBEDDING_DIM': CONFIG['EMBEDDING_DIM_NATIVE'],
'TRAINABLE': False
}}
print("Configuration adaptée pour EfficientNet")
print(f"Phase 1: extraction d'embedding simple ({CONFIG['EMBEDDING_DIM_NATIVE']}-dim)")
print("Imports et configuration OK!")
print(f"TensorFlow: {tf.__version__}")
print(f"GPU: {tf.config.list_physical_devices('GPU')}")
Configuration adaptée pour EfficientNet Phase 1: extraction d'embedding simple (1280-dim) Imports et configuration OK! TensorFlow: 2.19.0 GPU: []
2. Chargement du dataset¶
In [2]:
try:
import kagglehub
dataset_path = kagglehub.dataset_download("pes12017000148/food-ingredients-and-recipe-dataset-with-images")
print(f"Dataset téléchargé: {dataset_path}")
images_dir = os.path.join(dataset_path, "Food Images", "Food Images")
csv_files = [f for f in os.listdir(dataset_path) if f.endswith('.csv')]
csv_path = os.path.join(dataset_path, csv_files[0]) if csv_files else None
print(f"Images: {images_dir}")
print(f"CSV: {csv_path}")
except ImportError:
print("kagglehub non installé, veuillez l'installer avec: pip install kagglehub")
exit()
# Charger le CSV
df = pd.read_csv(csv_path)
print(f"{len(df)} recettes chargées")
print(f"Colonnes: {list(df.columns)}")
# Nettoyage basique
df_clean = df.dropna(subset=['Title', 'Ingredients', 'Instructions', 'Image_Name']).copy()
print(f"Après nettoyage: {len(df_clean)} recettes")
# Créer le texte complet
def clean_text(text):
if pd.isna(text):
return ""
text = str(text).lower().strip()
text = re.sub(r'[^a-zA-Z0-9\s,.!?-]', '', text)
text = re.sub(r'\s+', ' ', text)
return text
df_clean['full_text'] = (
df_clean['Title'].apply(clean_text) + ' ' +
df_clean['Ingredients'].apply(clean_text) + ' ' +
df_clean['Instructions'].apply(clean_text)
)
print("Données chargées")
Dataset téléchargé: C:\Users\Naturel\.cache\kagglehub\datasets\pes12017000148\food-ingredients-and-recipe-dataset-with-images\versions\1 Images: C:\Users\Naturel\.cache\kagglehub\datasets\pes12017000148\food-ingredients-and-recipe-dataset-with-images\versions\1\Food Images\Food Images CSV: C:\Users\Naturel\.cache\kagglehub\datasets\pes12017000148\food-ingredients-and-recipe-dataset-with-images\versions\1\Food Ingredients and Recipe Dataset with Image Name Mapping.csv 13501 recettes chargées Colonnes: ['Unnamed: 0', 'Title', 'Ingredients', 'Instructions', 'Image_Name', 'Cleaned_Ingredients'] Après nettoyage: 13493 recettes Données chargées
In [3]:
print("Vérification des images...")
if not os.path.exists(images_dir):
print(f"Dossier non trouvé: {images_dir}")
exit()
available_images = set(os.listdir(images_dir))
print(f"{len(available_images)} images trouvées")
print("- Exemples d'images:")
for i, img in enumerate(list(available_images)[:5]):
print(f"{img}")
print("- Exemples de noms CSV:")
for i, name in enumerate(df_clean['Image_Name'].dropna().head(5)):
print(f"{name}")
# Fonction de correspondance
def find_image_path(image_name):
"""Trouve le chemin complet de l'image"""
if pd.isna(image_name):
return None
# Essayer nom exact
if image_name in available_images:
return os.path.join(images_dir, image_name)
# Essayer avec .jpg
if f"{image_name}.jpg" in available_images:
return os.path.join(images_dir, f"{image_name}.jpg")
# Essayer avec .jpeg
if f"{image_name}.jpeg" in available_images:
return os.path.join(images_dir, f"{image_name}.jpeg")
# Essayer avec .png
if f"{image_name}.png" in available_images:
return os.path.join(images_dir, f"{image_name}.png")
return None
df_clean['image_path'] = df_clean['Image_Name'].apply(find_image_path)
df_clean['has_image'] = df_clean['image_path'].notna()
# Filtrer les recettes avec images
recipes_with_images = df_clean[df_clean['has_image']].copy()
print(f"\n{len(recipes_with_images)} recettes avec images conservées dans recipes_with_images")
Vérification des images... 13582 images trouvées - Exemples d'images: arugula-and-mint-salad-with-oil-cured-black-olives-oranges-and-ricotta-salata-236186.jpg watermelon-granita-with-gingered-strawberries-360529.jpg turkey-tetrazzini-13377.jpg cucumber-sake-tini-358951.jpg chawan-mushi-with-shrimp-and-spring-peas-51155250.jpg - Exemples de noms CSV: miso-butter-roast-chicken-acorn-squash-panzanella crispy-salt-and-pepper-potatoes-dan-kluger thanksgiving-mac-and-cheese-erick-williams italian-sausage-and-bread-stuffing-240559 newtons-law-apple-bourbon-cocktail 13463 recettes avec images conservées dans recipes_with_images
3. Nettoyage NLP¶
In [4]:
# Nettoyage NLP
import nltk
import pandas as pd
import re
# NLTK
print("Téléchargement des ressources NLTK...")
nltk.download('stopwords', quiet=True)
nltk.download('punkt_tab', quiet=True) # Version récente de punkt
nltk.download('wordnet', quiet=True)
nltk.download('omw-1.4', quiet=True) # Pour WordNet multilingue
print("Ressources NLTK OK!")
from nltk.corpus import stopwords
from nltk.tokenize import word_tokenize
from nltk.stem import WordNetLemmatizer
import string
# Configuration NLTK avec stop words culinaires personnalisés
stop_words = set(stopwords.words('english'))
# Stop words spécifiques à la cuisine, pourrait être amélioré
culinary_stopwords = {
'bake',
'baking',
'chopped',
'coarse',
'coarsely',
'cook',
'cooking',
'cube',
'cubes',
'cup',
'cups',
'cut',
'diced',
'divided',
'finely',
'gram',
'grams',
'grated',
'halved',
'hour',
'hours',
'inch',
'inches',
'kg',
'lengthwise',
'liter',
'liters',
'melted',
'minute',
'minutes',
'minced',
'ml',
'oz',
'peeled',
'piece',
'pieces',
'plus',
'pound',
'pounds',
'quartered',
'recipe',
'recipes',
'room',
'seeded',
'serve',
'serves',
'serving',
'servings',
'sliced',
'stick',
'tablespoon',
'tablespoons',
'tbsp',
'teaspoon',
'teaspoons',
'temperature',
'thinly',
'tsp',
'trimmed'
}
# Fusionner les stop words
stop_words.update(culinary_stopwords)
print(f"Stop words enrichis: {len(stop_words)} mots au total comprenant {len(culinary_stopwords)} stop words culinaires")
lemmatizer = WordNetLemmatizer()
def clean_text_advanced(text):
"""
Nettoyage avancé du texte avec NLTK
"""
if pd.isna(text) or text == '':
return ''
# Convertir en string et minuscules
text = str(text).lower()
# Supprimer la ponctuation et caractères spéciaux
text = text.translate(str.maketrans('', '', string.punctuation))
# Supprimer les chiffres
text = re.sub(r'\d+', '', text)
# Tokenisation
tokens = word_tokenize(text)
# Supprimer stopwords et mots courts (< 3 caractères)
tokens = [token for token in tokens if token not in stop_words and len(token) > 2]
# Lemmatisation
tokens = [lemmatizer.lemmatize(token) for token in tokens]
return ' '.join(tokens)
def extract_ingredients_list(ingredients_str):
"""
Extrait et nettoie la liste d'ingrédients depuis le format string
"""
if pd.isna(ingredients_str):
return []
# Supprimer les crochets et guillemets
cleaned = str(ingredients_str).strip("[]'\"")
# Séparer par virgules et nettoyer
ingredients = []
for item in cleaned.split(','):
item = item.strip("' \"")
if item and len(item) > 1:
# Nettoyer chaque ingrédient
clean_ingredient = clean_text_advanced(item)
if clean_ingredient:
ingredients.append(clean_ingredient)
return ingredients
# Nettoyage NLP
print("Application du nettoyage NLP...")
recipes_with_images['Title_Clean'] = recipes_with_images['Title'].apply(clean_text_advanced)
recipes_with_images['Ingredients_Clean'] = recipes_with_images['Ingredients'].apply(clean_text_advanced)
recipes_with_images['Instructions_Clean'] = recipes_with_images['Instructions'].apply(clean_text_advanced)
# Extraire la liste des ingrédients
recipes_with_images['Ingredients_List'] = recipes_with_images['Ingredients'].apply(extract_ingredients_list)
# Recréer le texte complet avec le nettoyage avancé
recipes_with_images['full_text_advanced'] = (
recipes_with_images['Title_Clean'] + ' ' +
recipes_with_images['Ingredients_Clean'] + ' ' +
recipes_with_images['Instructions_Clean']
)
print("Nettoyage NLP terminé!")
Téléchargement des ressources NLTK... Ressources NLTK OK! Stop words enrichis: 257 mots au total comprenant 59 stop words culinaires Application du nettoyage NLP... Nettoyage NLP terminé!
4. Visualisation des images du dataset¶
In [5]:
def visualize_recipe_gallery_harmonized(recipes_df, images_dir, grid_size=(4, 3), random_seed=42):
"""
Affiche une galerie d'images de recettes
"""
# Configuration du style
plt.style.use('dark_background')
# Créer la figure avec le fond sombre harmonisé
fig, axes = plt.subplots(grid_size[0], grid_size[1], figsize=(18, 16))
fig.patch.set_facecolor('#1a1a1a') # Même fond que les autres visualisations
# Titre principal avec style harmonisé
fig.suptitle('GALERIE DE RECETTES',
fontsize=24, fontweight='bold', color='white', y=0.98)
# Aplatir les axes pour faciliter l'itération
axes = axes.flatten()
# Échantillonner des recettes aléatoirement
total_samples = grid_size[0] * grid_size[1]
if len(recipes_df) >= total_samples:
sample_recipes = recipes_df.sample(n=total_samples, random_state=random_seed)
else:
sample_recipes = recipes_df.sample(n=len(recipes_df), random_state=random_seed)
print(f"Seulement {len(recipes_df)} recettes disponibles (demandé: {total_samples})")
# Palette de couleurs pour les bordures
border_colors = ['#FF6B9D', '#4ECDC4', '#45B7D1', '#96CEB4', '#FECA57',
'#FF9FF3', '#54A0FF', '#5F27CD', '#A8E6CF', '#FFD93D']
for idx, (_, recipe) in enumerate(sample_recipes.iterrows()):
if idx >= len(axes):
break
ax = axes[idx]
ax.set_facecolor('#2d2d2d')
try:
# Utiliser le chemin d'image déjà résolu
image_path = recipe.get('image_path')
if image_path and os.path.exists(image_path):
# Charger et afficher l'image
img = Image.open(image_path)
img = img.convert('RGB')
ax.imshow(img)
# Titre tronqué si trop long
title = recipe['Title']
if len(title) > 35:
title = title[:35] + "..."
border_color = border_colors[idx % len(border_colors)]
ax.set_title(title, fontsize=11, fontweight='bold',
color='white', pad=15,
bbox=dict(boxstyle="round,pad=0.5",
facecolor='#1a1a1a',
edgecolor=border_color,
linewidth=2, alpha=0.9))
ax.axis('off')
# Ajouter une bordure colorée autour de l'image
for spine in ax.spines.values():
spine.set_visible(True)
spine.set_color(border_color)
spine.set_linewidth(3)
spine.set_alpha(0.8)
else:
# Placeholder pour images manquantes
ax.text(0.5, 0.5, f'\n\nImage non trouvée\n\n"{recipe["Title"][:25]}..."',
horizontalalignment='center', verticalalignment='center',
transform=ax.transAxes, fontsize=10, color='white',
bbox=dict(boxstyle="round,pad=0.8",
facecolor='#2d2d2d',
edgecolor='#FF6B9D',
linewidth=2, alpha=0.9))
ax.set_title("Image manquante", fontsize=11, color='#FF6B9D', fontweight='bold')
ax.axis('off')
except Exception as e:
ax.text(0.5, 0.5, f'\n\nErreur de chargement\n\n{str(e)[:20]}...',
horizontalalignment='center', verticalalignment='center',
transform=ax.transAxes, fontsize=10, color='white',
bbox=dict(boxstyle="round,pad=0.8",
facecolor='#2d2d2d',
edgecolor='#FECA57',
linewidth=2, alpha=0.9))
ax.set_title("Erreur", fontsize=11, color='#FECA57', fontweight='bold')
ax.axis('off')
# Masquer des axes non utilisés
for idx in range(len(sample_recipes), len(axes)):
axes[idx].set_facecolor('#1a1a1a')
axes[idx].axis('off')
plt.tight_layout(pad=3.0, rect=[0, 0.03, 1, 0.96])
plt.show()
# Statistiques
print(f"\nSTATISTIQUES SUR LES IMAGES:")
print(f"Images affichées: {min(len(sample_recipes), len(axes))}")
print(f"Total recettes disponibles: {len(recipes_df):,}")
print(f"Recettes avec images: {recipes_df['has_image'].sum():,}")
print(f"Pourcentage avec images: {(recipes_df['has_image'].sum() / len(recipes_df) * 100):.1f}%")
# Afficher quelques exemples de titres
print(f"\nEXEMPLES DE TITRES:")
for i, (_, recipe) in enumerate(sample_recipes.head(3).iterrows()):
print(f"{i+1}. {recipe['Title']}")
# Affichage de la galerie
if 'recipes_with_images' in globals() and 'images_dir' in globals():
if len(recipes_with_images) > 0:
visualize_recipe_gallery_harmonized(recipes_with_images, images_dir, grid_size=(4, 3))
else:
print("Aucune recette avec image trouvée pour la galerie")
else:
print("Erreur. Assurez-vous d'avoir exécuté les cellules de chargement des données.")
if 'recipes_with_images' not in globals():
print("- 'recipes_with_images' non trouvé")
if 'images_dir' not in globals():
print("- 'images_dir' non trouvé")
STATISTIQUES SUR LES IMAGES: Images affichées: 12 Total recettes disponibles: 13,463 Recettes avec images: 13,463 Pourcentage avec images: 100.0% EXEMPLES DE TITRES: 1. Smoked Salmon, Fennel and Goat Cheese Toasts 2. Spicy Thai Tofu with Red Bell Peppers and Peanuts 3. Steak Salad with Caraway Vinaigrette and Rye Croutons
5. Analyse exploratoire des données (recipes_with_images)¶
In [6]:
import platform
from wordcloud import WordCloud
# Configuration dark theme
plt.style.use('dark_background')
plt.rcParams['font.family'] = ['Segoe UI Emoji', 'Apple Color Emoji', 'Noto Color Emoji', 'DejaVu Sans', 'sans-serif']
plt.rcParams['axes.unicode_minus'] = False
if platform.system() == "Windows":
plt.rcParams['font.family'] = ['Segoe UI Emoji', 'DejaVu Sans', 'sans-serif']
elif platform.system() == "Linux":
plt.rcParams['font.family'] = ['Noto Color Emoji', 'DejaVu Sans', 'Liberation Sans', 'sans-serif']
elif platform.system() == "Darwin": # macOS
plt.rcParams['font.family'] = ['Apple Color Emoji', 'SF Pro Display', 'DejaVu Sans', 'sans-serif']
else:
plt.rcParams['font.family'] = ['DejaVu Sans', 'sans-serif']
plt.rcParams['axes.unicode_minus'] = False
warnings.filterwarnings('ignore', category=UserWarning, module='matplotlib.font_manager')
# Palette de couleurs magnifique (IDENTIQUE)
beautiful_colors = ['#FF6B9D', '#4ECDC4', '#45B7D1', '#96CEB4', '#FECA57', '#FF9FF3', '#54A0FF', '#5F27CD', '#A8E6CF', '#FFD93D']
def create_wordcloud_harmonized(text_data, title, max_words=100, colormap='viridis'):
"""
Crée un nuage de mots
"""
# Combiner tout le texte
if isinstance(text_data, list):
combined_text = ' '.join([str(text) for text in text_data if pd.notna(text) and str(text).strip()])
else:
combined_text = ' '.join([str(text) for text in text_data.dropna() if str(text).strip()])
if not combined_text.strip():
print(f" Pas de données texte pour {title}")
return None
# Créer le WordCloud avec le style harmonisé
wordcloud = WordCloud(
width=800,
height=400,
background_color='#2d2d2d', # Même fond que les autres graphiques
max_words=max_words,
colormap=colormap,
collocations=False,
prefer_horizontal=0.7
).generate(combined_text)
return wordcloud
def plot_top_ingredients_harmonized(recipes_df, ax, top_n=20):
"""
Graphique des ingrédients
"""
ax.set_facecolor('#2d2d2d') # Même fond
# Compter tous les ingrédients
all_ingredients = []
if 'Ingredients_List' in recipes_df.columns:
for ingredients_list in recipes_df['Ingredients_List'].dropna():
if isinstance(ingredients_list, list):
all_ingredients.extend(ingredients_list)
else:
# Fallback: extraire depuis la colonne Ingredients
for ingredients_text in recipes_df['Ingredients'].dropna():
if pd.notna(ingredients_text) and str(ingredients_text).strip():
ingredients = [ing.strip().lower() for ing in str(ingredients_text).split(',')]
all_ingredients.extend([ing for ing in ingredients if len(ing) > 2])
if not all_ingredients:
ax.text(0.5, 0.5, " Aucun ingrédient trouvé", ha='center', va='center',
transform=ax.transAxes, color='white', fontsize=12)
return None
# Compter les fréquences
from collections import Counter
ingredient_counts = Counter(all_ingredients)
top_ingredients = ingredient_counts.most_common(top_n)
if not top_ingredients:
return None
# Créer le graphique avec le style harmonisé
ingredients, counts = zip(*top_ingredients)
# Utiliser la palette magnifique
colors = [beautiful_colors[i % len(beautiful_colors)] for i in range(len(ingredients))]
bars = ax.barh(range(len(ingredients)), counts,
color=colors, alpha=0.9,
edgecolor='white', linewidth=0.8) # Bordures plus fines
ax.set_xlabel('Fréquence', fontweight='bold', color='white', fontsize=12)
ax.set_ylabel('Ingrédients', fontweight='bold', color='white', fontsize=12)
ax.set_title(f' Top {top_n} Ingrédients', fontweight='bold', fontsize=16, color='white', pad=20)
ax.set_yticks(range(len(ingredients)))
ax.set_yticklabels(ingredients, fontsize=10, color='white')
ax.tick_params(colors='white')
# Ajouter les valeurs avec style
for bar, count in zip(bars, counts):
ax.text(bar.get_width() + max(counts)*0.02, bar.get_y() + bar.get_height()/2,
str(count), ha='left', va='center', fontweight='bold',
color='white', fontsize=11)
# Style harmonisé
ax.invert_yaxis()
ax.grid(True, alpha=0.3, color='#404040', linestyle='--')
ax.spines['bottom'].set_color('#404040')
ax.spines['top'].set_color('#404040')
ax.spines['right'].set_color('#404040')
ax.spines['left'].set_color('#404040')
return top_ingredients
def plot_instruction_lengths_harmonized(recipes_df, ax):
"""
Distribution des longueurs des instructions
"""
ax.set_facecolor('#2d2d2d')
instruction_lengths = recipes_df['Instructions'].dropna().apply(lambda x: len(str(x).split()))
# Histogramme
n, bins, patches = ax.hist(instruction_lengths, bins=30, alpha=0.9,
edgecolor='white', linewidth=0.5)
base_color = np.array([69, 183, 209])
for i, patch in enumerate(patches):
intensity = 0.4 + 0.6 * (i / len(patches))
color = base_color * intensity / 255.0
patch.set_facecolor(color)
ax.set_xlabel('Nombre de mots dans les instructions', fontweight='bold', color='white', fontsize=12)
ax.set_ylabel('Fréquence', fontweight='bold', color='white', fontsize=12)
ax.set_title('Distribution des Instructions', fontweight='bold', fontsize=16, color='white', pad=20)
ax.tick_params(colors='white')
# Statistiques avec couleurs contrastantes pour la lisibilité
mean_length = instruction_lengths.mean()
median_length = instruction_lengths.median()
ax.axvline(mean_length, color='#FFFFFF', linestyle='--', linewidth=3, alpha=0.9, label=f' Moyenne: {mean_length:.1f}')
ax.axvline(median_length, color='#FFD700', linestyle='--', linewidth=3, alpha=0.9, label=f' Médiane: {median_length:.1f}')
# Légende harmonisée
legend = ax.legend(framealpha=0.9, facecolor='#2d2d2d', edgecolor='white')
for text in legend.get_texts():
text.set_color('white')
ax.grid(True, alpha=0.3, color='#404040', linestyle='--')
ax.spines['bottom'].set_color('#404040')
ax.spines['top'].set_color('#404040')
ax.spines['right'].set_color('#404040')
ax.spines['left'].set_color('#404040')
return instruction_lengths
if 'recipes_with_images' in globals():
fig = plt.figure(figsize=(20, 14))
fig.patch.set_facecolor('#1a1a1a')
fig.suptitle('ANALYSE DU DATASET AVEC IMAGES',
fontsize=24, fontweight='bold', color='white', y=0.98)
# WordCloud des ingrédients
ax1 = plt.subplot(2, 2, 1)
ax1.set_facecolor('#2d2d2d')
if 'Ingredients_Clean' in recipes_with_images.columns:
ingredients_text = recipes_with_images['Ingredients_Clean']
else:
ingredients_text = recipes_with_images['Ingredients']
wordcloud_ingredients = create_wordcloud_harmonized(ingredients_text, "Ingrédients", max_words=120, colormap='plasma')
if wordcloud_ingredients:
ax1.imshow(wordcloud_ingredients, interpolation='bilinear')
ax1.set_title('Nuage de mots - Ingrédients', fontweight='bold', fontsize=16, color='#FF6B9D', pad=20)
ax1.axis('off')
else:
ax1.text(0.5, 0.5, "Pas de données d'ingrédients", ha='center', va='center',
transform=ax1.transAxes, color='white', fontsize=14)
# WordCloud des titres
ax2 = plt.subplot(2, 2, 2)
ax2.set_facecolor('#2d2d2d')
if 'Title_Clean' in recipes_with_images.columns:
titles_text = recipes_with_images['Title_Clean']
else:
titles_text = recipes_with_images['Title']
wordcloud_titles = create_wordcloud_harmonized(titles_text, "Titres", max_words=120, colormap='viridis')
if wordcloud_titles:
ax2.imshow(wordcloud_titles, interpolation='bilinear')
ax2.set_title('Nuage de mots - Titres de Recettes', fontweight='bold', fontsize=16, color='#4ECDC4', pad=20)
ax2.axis('off')
else:
ax2.text(0.5, 0.5, "Pas de données de titres", ha='center', va='center',
transform=ax2.transAxes, color='white', fontsize=14)
# Top 20 ingrédients
ax3 = plt.subplot(2, 2, 3)
print("Analyse des top 20 ingrédients...")
top_ingredients = plot_top_ingredients_harmonized(recipes_with_images, ax3, top_n=20)
if top_ingredients:
print(f"Top 20 ingrédients: {[ing for ing, count in top_ingredients[:20]]}")
# Distribution des instructions
ax4 = plt.subplot(2, 2, 4)
instruction_lengths = plot_instruction_lengths_harmonized(recipes_with_images, ax4)
plt.tight_layout(pad=4.0, rect=[0, 0.03, 1, 1])
plt.show()
# Statistiques
print(f"\nSTATISTIQUES DES INSTRUCTIONS:")
print(f"Longueur moyenne: {instruction_lengths.mean():.1f} mots")
print(f"Longueur médiane: {instruction_lengths.median():.1f} mots")
print(f"Plus courte: {instruction_lengths.min()} mots")
print(f"Plus longue: {instruction_lengths.max()} mots")
print(f"\nSTATISTIQUES GÉNÉRALES:")
print(f"Recettes avec images: {len(recipes_with_images):,}")
print(f"Recettes avec ingrédients: {recipes_with_images['Ingredients'].notna().sum():,}")
print(f"Recettes avec instructions: {recipes_with_images['Instructions'].notna().sum():,}")
print(f"Recettes avec nettoyage avancé: {recipes_with_images['full_text_advanced'].notna().sum() if 'full_text_advanced' in recipes_with_images.columns else 'N/A'}")
# ANALYSE DES DOUBLONS
def clean_text_for_comparison(text):
"""Nettoie le texte pour la comparaison des doublons"""
if pd.isna(text) or text is None:
return ""
return str(text).lower().strip().replace('\n', ' ').replace('\r', ' ')
print(f"\nANALYSE DES TITRES EN DOUBLE:")
recipes_with_images['title_clean_comparison'] = recipes_with_images['Title'].apply(clean_text_for_comparison)
# Compter les doublons de titres
title_counts = recipes_with_images['title_clean_comparison'].value_counts()
duplicate_titles = title_counts[title_counts > 1]
unique_titles = title_counts[title_counts == 1]
print(f"Titres uniques: {len(unique_titles):,}")
print(f"Titres en double: {len(duplicate_titles):,}")
print(f"Total d'occurrences dupliquées: {duplicate_titles.sum():,}")
print(f"Pourcentage de doublons: {(len(duplicate_titles) / len(title_counts) * 100):.1f}%")
# Analyse des instructions en double
print(f"\nANALYSE DES INSTRUCTIONS EN DOUBLE:")
recipes_with_images['instructions_clean_comparison'] = recipes_with_images['Instructions'].apply(clean_text_for_comparison)
# Filtrer les instructions vides
non_empty_instructions = recipes_with_images[recipes_with_images['instructions_clean_comparison'].str.len() > 10]
instruction_counts = non_empty_instructions['instructions_clean_comparison'].value_counts()
duplicate_instructions = instruction_counts[instruction_counts > 1]
unique_instructions = instruction_counts[instruction_counts == 1]
print(f"Instructions uniques: {len(unique_instructions):,}")
print(f"Instructions en double: {len(duplicate_instructions):,}")
print(f"Total d'occurrences dupliquées: {duplicate_instructions.sum():,}")
print(f"Pourcentage de doublons: {(len(duplicate_instructions) / len(instruction_counts) * 100):.1f}%")
# Analyse des recettes complètement identiques (titre + instructions)
print(f"\nANALYSE DES RECETTES COMPLÈTEMENT IDENTIQUES:")
recipes_with_images['combined_clean'] = (
recipes_with_images['title_clean_comparison'] + " ||| " +
recipes_with_images['instructions_clean_comparison']
)
combined_counts = recipes_with_images['combined_clean'].value_counts()
identical_recipes = combined_counts[combined_counts > 1]
print(f"Recettes complètement uniques: {len(combined_counts[combined_counts == 1]):,}")
print(f"Recettes complètement identiques: {len(identical_recipes):,}")
print(f"Total de recettes dupliquées: {identical_recipes.sum():,}")
# VISUALISATION DES DOUBLONS
fig_duplicates = plt.figure(figsize=(20, 16))
fig_duplicates.patch.set_facecolor('#1a1a1a')
fig_duplicates.suptitle(' ANALYSE VISUELLE DES DOUBLONS',
fontsize=22, fontweight='bold', color='white', y=0.98)
# Graphique en secteurs pour les recettes uniques vs dupliquées
ax_pie = plt.subplot(3, 3, 1)
ax_pie.set_facecolor('#2d2d2d')
unique_count = len(combined_counts[combined_counts == 1])
duplicate_count = identical_recipes.sum()
sizes = [unique_count, duplicate_count]
labels = [f'Uniques\n({unique_count:,})', f' Dupliquées\n({duplicate_count:,})']
colors = ['#4ECDC4', '#FF6B9D']
explode = (0.05, 0.1)
wedges, texts, autotexts = ax_pie.pie(sizes, labels=labels, colors=colors,
autopct='%1.1f%%', explode=explode,
shadow=True, startangle=90,
textprops={'fontsize': 10, 'color': 'white', 'fontweight': 'bold'})
ax_pie.set_title('Uniques vs Dupliquées',
fontweight='bold', fontsize=14, color='white', pad=15)
# Graphique en barres des types de doublons
ax_bar = plt.subplot(3, 3, 2)
ax_bar.set_facecolor('#2d2d2d')
duplicate_types = ['Titres', 'Instructions', 'Complètes']
duplicate_counts_values = [
len(duplicate_titles),
len(duplicate_instructions),
len(identical_recipes)
]
bars = ax_bar.bar(duplicate_types, duplicate_counts_values,
color=['#FF9FF3', '#54A0FF', '#FECA57'],
alpha=0.9, edgecolor='white', linewidth=1.5)
ax_bar.set_title(' Types de Doublons', fontweight='bold', fontsize=14, color='white', pad=15)
ax_bar.set_ylabel('Nombre', fontweight='bold', color='white', fontsize=11)
ax_bar.tick_params(colors='white', labelsize=10)
# Ajouter les valeurs sur les barres
for bar, count in zip(bars, duplicate_counts_values):
ax_bar.text(bar.get_x() + bar.get_width()/2, bar.get_height() + max(duplicate_counts_values)*0.02,
str(count), ha='center', va='bottom', fontweight='bold',
color='white', fontsize=11)
ax_bar.grid(True, alpha=0.3, color='#404040', linestyle='--')
for spine in ax_bar.spines.values():
spine.set_color('#404040')
# 3. Texte d'information
ax_info = plt.subplot(3, 3, 3)
ax_info.set_facecolor('#2d2d2d')
ax_info.axis('off')
info_text = f"""STATISTIQUES DOUBLONS
Total recettes: {len(recipes_with_images):,}
Titres dupliqués: {len(duplicate_titles):,}
Instructions dupliquées: {len(duplicate_instructions):,}
Recettes identiques: {len(identical_recipes):,}
Pourcentage de doublons:
- Titres: {(len(duplicate_titles) / len(title_counts) * 100):.1f}%
- Instructions: {(len(duplicate_instructions) / len(instruction_counts) * 100):.1f}%"""
ax_info.text(0.05, 0.95, info_text, transform=ax_info.transAxes,
fontsize=11, color='white', va='top', ha='left',
bbox=dict(boxstyle="round,pad=0.3", facecolor='#404040', alpha=0.8))
# 4-7. Exemple de recette dupliquée avec images (4 sous-plots pour les images)
if len(identical_recipes) > 0:
# Prendre le premier doublon avec le plus d'occurrences
most_duplicated = identical_recipes.head(1)
duplicate_key = most_duplicated.index[0]
duplicate_count_example = most_duplicated.iloc[0]
# Trouver toutes les recettes identiques
duplicate_recipes = recipes_with_images[recipes_with_images['combined_clean'] == duplicate_key]
# Titre de l'exemple
title_part = duplicate_key.split(" ||| ")[0]
# Informations de l'exemple
ax_example_info = plt.subplot(3, 1, 2)
ax_example_info.set_facecolor('#2d2d2d')
ax_example_info.axis('off')
ax_example_info.text(0.5, 0.9, f'EXEMPLE DE RECETTE DUPLIQUÉE ({duplicate_count_example} fois)',
ha='center', va='top', transform=ax_example_info.transAxes,
fontsize=16, fontweight='bold', color='#FFD93D')
ax_example_info.text(0.5, 0.7, f'Titre: "{title_part}"',
ha='center', va='top', transform=ax_example_info.transAxes,
fontsize=12, fontweight='bold', color='white')
# Instructions preview
instructions_part = duplicate_key.split(" ||| ")[1]
preview_instructions = instructions_part[:150] + "..." if len(instructions_part) > 150 else instructions_part
ax_example_info.text(0.5, 0.4, f'Instructions: "{preview_instructions}"',
ha='center', va='top', transform=ax_example_info.transAxes,
fontsize=10, color='#A8E6CF', style='italic')
# Images des recettes dupliquées (4 subplots)
valid_images = []
for idx, recipe in duplicate_recipes.head(4).iterrows():
if 'image_path' in recipe and pd.notna(recipe['image_path']):
try:
img_path = recipe['image_path']
if os.path.exists(img_path):
valid_images.append((img_path, recipe['Title']))
except:
pass
# Positions pour les 4 images en bas
image_positions = [(3, 4, 9), (3, 4, 10), (3, 4, 11), (3, 4, 12)] # 3 rows, 4 cols, positions 9-12
for i in range(4):
ax_img = plt.subplot(*image_positions[i])
ax_img.set_facecolor('#2d2d2d')
if i < len(valid_images):
try:
img_path, original_title = valid_images[i]
img = plt.imread(img_path)
ax_img.imshow(img)
ax_img.set_title(f'Copie #{i+1}', fontsize=11, color='white', fontweight='bold', pad=10)
# Titre sous l'image
title_display = original_title[:25] + "..." if len(original_title) > 25 else original_title
ax_img.text(0.5, -0.15, title_display, transform=ax_img.transAxes,
ha='center', va='top', fontsize=9, color='#96CEB4', style='italic')
except Exception as e:
ax_img.text(0.5, 0.5, f'\nErreur image #{i+1}', ha='center', va='center',
transform=ax_img.transAxes, fontsize=10, color='#FF6B9D')
else:
ax_img.text(0.5, 0.5, f'\nPas d\'image #{i+1}', ha='center', va='center',
transform=ax_img.transAxes, fontsize=10, color='#888888')
ax_img.axis('off')
else:
# Pas de doublons complets
ax_no_duplicates = plt.subplot(3, 1, (2, 3))
ax_no_duplicates.set_facecolor('#2d2d2d')
ax_no_duplicates.axis('off')
ax_no_duplicates.text(0.5, 0.5, 'Aucun doublon complet trouvé!\nVotre dataset est très propre!',
ha='center', va='center', transform=ax_no_duplicates.transAxes,
fontsize=18, color='#4ECDC4', fontweight='bold')
plt.tight_layout(pad=2.0, rect=[0, 0.02, 1, 0.98])
plt.show()
# Nettoyage des colonnes temporaires
recipes_with_images.drop(['title_clean_comparison', 'instructions_clean_comparison', 'combined_clean'],
axis=1, inplace=True)
print("Toutes les visualisations et l'analyse des doublons sont OK!")
else:
print("DataFrame 'recipes_with_images' non trouvé. Exécutez d'abord les cellules de chargement et de correspondance des images.")
Analyse des top 20 ingrédients... Top 20 ingrédients: ['kosher salt', 'sugar', 'salt', 'unsalted butter', 'olive oil', 'garlic clove', 'large egg', 'extravirgin olive oil', 'freshly ground black pepper', 'fresh lemon juice', 'allpurpose flour', 'water', 'vegetable oil', 'vanilla extract', 'powder', 'whole milk', 'freshly ground pepper', 'heavy cream', 'honey', 'fresh lime juice']
STATISTIQUES DES INSTRUCTIONS: Longueur moyenne: 180.9 mots Longueur médiane: 153.0 mots Plus courte: 6 mots Plus longue: 2587 mots STATISTIQUES GÉNÉRALES: Recettes avec images: 13,463 Recettes avec ingrédients: 13,463 Recettes avec instructions: 13,463 Recettes avec nettoyage avancé: 13463 ANALYSE DES TITRES EN DOUBLE: Titres uniques: 13,104 Titres en double: 164 Total d'occurrences dupliquées: 359 Pourcentage de doublons: 1.2% ANALYSE DES INSTRUCTIONS EN DOUBLE: Instructions uniques: 13,419 Instructions en double: 15 Total d'occurrences dupliquées: 44 Pourcentage de doublons: 0.1% ANALYSE DES RECETTES COMPLÈTEMENT IDENTIQUES: Recettes complètement uniques: 13,441 Recettes complètement identiques: 8 Total de recettes dupliquées: 22
Toutes les visualisations et l'analyse des doublons sont OK!
6. Définition du modèle (EfficientNet gelé)¶
In [7]:
# Modèle => EfficientNet
from sklearn.metrics.pairwise import cosine_similarity
from PIL import Image
def create_image_retrieval_model(embedding_dim=1280):
"""
Create EfficientNet model for image embedding extraction
"""
# Load pre-trained EfficientNet
base_model = EfficientNetB0(
weights='imagenet',
include_top=False,
input_shape=(CONFIG['IMG_SIZE'], CONFIG['IMG_SIZE'], 3)
)
# Input shape
inputs = Input(shape=(CONFIG['IMG_SIZE'], CONFIG['IMG_SIZE'], 3))
x = base_model(inputs, training=False)
x = GlobalAveragePooling2D()(x)
# Normalisation L2 (mauvaise pratique de faire comme ça)
def l2_norm(x):
return tf.nn.l2_normalize(x, axis=1)
# Same shape
def same_shape(input_shape):
return input_shape
# Voir recipe_image_retrieval_tl.ipynb on a fait mieux avec des custom layers car pour charger le modèle c'est contraignant de faire comme ça
embeddings = Lambda(l2_norm, output_shape=same_shape, name="l2_norm")(x)
model = Model(inputs, embeddings, name='image_retrieval_model')
return model
# Modèle créé
image_retrieval_model = create_image_retrieval_model(CONFIG['EMBEDDING_DIM_NATIVE'])
print("Modèle chargé!")
print(f"Paramètres: {image_retrieval_model.count_params():,}")
print(f"Dimension des embeddings: {CONFIG['EMBEDDING_DIM_NATIVE']}")
# Affiché le modèle simple
image_retrieval_model.summary()
Modèle chargé! Paramètres: 4,049,571 Dimension des embeddings: 1280
Model: "image_retrieval_model"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓ ┃ Layer (type) ┃ Output Shape ┃ Param # ┃ ┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩ │ input_layer_1 (InputLayer) │ (None, 224, 224, 3) │ 0 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ efficientnetb0 (Functional) │ (None, 7, 7, 1280) │ 4,049,571 │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ global_average_pooling2d │ (None, 1280) │ 0 │ │ (GlobalAveragePooling2D) │ │ │ ├─────────────────────────────────┼────────────────────────┼───────────────┤ │ l2_norm (Lambda) │ (None, 1280) │ 0 │ └─────────────────────────────────┴────────────────────────┴───────────────┘
Total params: 4,049,571 (15.45 MB)
Trainable params: 4,007,548 (15.29 MB)
Non-trainable params: 42,023 (164.16 KB)
7. Création de la classe "Recipe Image Retrieval"¶
In [8]:
class RecipeImageRetrieval:
"""
Système complet pour trouver les images de recettes similaires
Input: Query image → Output: Top 3 images similaires + infos de la recette
Utilise un EfficientNetB0 pour l'extraction d'embedding (pas d'entraînement)
"""
def __init__(self, model, recipes_df):
self.model = model
self.recipes_df = recipes_df
self.embeddings_db = None
self.image_paths = None
self.image_to_recipe_map = {}
def prepare_database(self):
"""Récupération des images de recipes_with_images"""
print("On prépare la base de données des images de recipes_with_images...")
valid_recipes = self.recipes_df[self.recipes_df['image_path'].notna()].copy()
self.image_paths = valid_recipes['image_path'].tolist()
print(f"{len(self.image_paths)} images trouvées")
# Create mapping from image path to recipe info
for idx, row in valid_recipes.iterrows():
self.image_to_recipe_map[row['image_path']] = {
'title': row['Title'],
'ingredients': row['Ingredients'],
'instructions': row['Instructions'],
'original_index': idx
}
print("Mapping image-to-recipe créé!")
def build_embeddings_database(self, batch_size=32):
"""Extraction des embeddings de toutes les images de la base de données"""
if self.image_paths is None:
self.prepare_database()
print("Extraction des embeddings...")
embeddings = []
total_batches = len(self.image_paths) // batch_size + (1 if len(self.image_paths) % batch_size else 0)
for i in range(0, len(self.image_paths), batch_size):
batch_paths = self.image_paths[i:i+batch_size]
batch_images = []
print(f"Batch {i//batch_size + 1}/{total_batches}...")
for img_path in batch_paths:
try:
img = tf.keras.preprocessing.image.load_img(
img_path, target_size=(CONFIG['IMG_SIZE'], CONFIG['IMG_SIZE'])
)
img = tf.keras.preprocessing.image.img_to_array(img)
img = tf.keras.applications.efficientnet.preprocess_input(img)
batch_images.append(img)
except Exception as e:
print(f"Erreur chargement {img_path}: {e}")
batch_images.append(np.zeros((CONFIG['IMG_SIZE'], CONFIG['IMG_SIZE'], 3)))
if batch_images:
batch_images = np.array(batch_images)
batch_embeddings = self.model.predict(batch_images, verbose=0)
embeddings.extend(batch_embeddings)
self.embeddings_db = np.array(embeddings)
print(f"Base de données des embeddings créée: {len(self.embeddings_db)} embeddings")
print(f"Shape embeddings: {self.embeddings_db.shape}")
def visualize_model_architecture(self):
"""Visualisation de l'architecture du modèle"""
fig = plt.figure(figsize=(16, 12))
fig.patch.set_facecolor('#1a1a1a')
fig.suptitle('Architecture du modèle, extraction d\'embedding (pas d\'entraînement)',
fontsize=18, color='white', fontweight='bold', y=0.94)
total_params = self.model.count_params()
trainable_params = sum(layer.count_params() for layer in self.model.layers if layer.trainable)
frozen_params = total_params - trainable_params
total_recipes = len(self.recipes_df)
recipes_with_images = len(self.recipes_df[self.recipes_df['image_path'].notna()]) if 'image_path' in self.recipes_df.columns else 0
embeddings_ready = len(self.embeddings_db) if self.embeddings_db is not None else 0
# Distribution des paramètres (gelés et trainables), ici 100% gelés car pas d'entraînement
ax1 = plt.subplot(2, 2, 1)
ax1.set_facecolor('#2d2d2d')
params_data = [frozen_params, trainable_params] if trainable_params > 0 else [total_params]
params_labels = ['Gelés (pré-entraînés)', 'Entraînables'] if trainable_params > 0 else ['Tous gelés']
colors_params = ['#4ECDC4', '#FF6B9D'] if trainable_params > 0 else ['#4ECDC4']
wedges, texts, autotexts = ax1.pie(params_data, labels=params_labels, colors=colors_params,
autopct='%1.1f%%', startangle=90)
for text in texts:
text.set_color('white')
for autotext in autotexts:
autotext.set_color('white')
autotext.set_fontweight('bold')
ax1.set_title(f'Paramètres (pas d\'entraînement)\nTotal: {total_params:,}',
fontsize=14, color='white', fontweight='bold', pad=15)
# Pipeline d'extraction d'embedding
ax2 = plt.subplot(2, 2, 2)
ax2.set_facecolor('#2d2d2d')
ax2.set_xlim(0, 12)
ax2.set_ylim(0, 8)
pipeline_steps = [
(1.5, 5, "Image\n224x224x3", '#FF6B9D', 1.5),
(3.8, 5, "EfficientNet\n(Gelé)", '#4ECDC4', 2.0),
(6.5, 5, "Global Pool\n(1280-dim)", '#45B7D1', 1.8),
(9.5, 5, "L2 Normalize\n", '#96CEB4', 1.8)
]
# Dessiner les étapes
for i, (x, y, text, color, width) in enumerate(pipeline_steps):
ax2.add_patch(plt.Rectangle((x-width/2, y-0.9), width, 1.8,
facecolor=color, alpha=0.8, edgecolor='white', linewidth=2))
ax2.text(x, y, text, ha='center', va='center', fontsize=10,
color='white', fontweight='bold')
if i < len(pipeline_steps) - 1:
next_x = pipeline_steps[i+1][0]
next_width = pipeline_steps[i+1][4]
ax2.arrow(x + width/2 + 0.1, y, next_x - x - width/2 - next_width/2 - 0.2, 0,
head_width=0.3, head_length=0.2, fc='white', ec='white', alpha=0.8)
# Workflow explicatif
ax2.add_patch(plt.Rectangle((0.5, 2.3), 10.5, 1.8,
facecolor='#1a1a1a', alpha=0.9, edgecolor='#45B7D1', linewidth=2))
workflow_text = """
EXTRACTION DES EMBEDDINGS: 1x ce pipeline → Extraction de 1280-dim features
RECHERCHE: par Similarité Cosine entre l'input et la BDD des embeddings
"""
ax2.text(5.75, 3.2, workflow_text, ha='center', va='center', fontsize=10,
color='white', fontweight='bold')
ax2.set_title('Pipeline d\'extraction d\'embedding', fontsize=14, color='white', fontweight='bold', pad=15)
ax2.text(5.75, 0.8, "Pas d'entraînement - extraction d'embeddings en utilisant les poids ImageNet",
ha='center', va='center', fontsize=9, color='#45B7D1', style='italic')
ax2.axis('off')
# Statistiques de la BDD des embeddings
ax3 = plt.subplot(2, 2, 3)
ax3.set_facecolor('#2d2d2d')
database_stats = {
'Total\nrecettes': total_recipes,
'Avec\nimages': recipes_with_images,
'Embeddings\n': embeddings_ready,
'\nDim des embeddings': CONFIG['EMBEDDING_DIM_NATIVE']
}
bars = ax3.bar(list(database_stats.keys()), list(database_stats.values()),
color=['#FF6B9D', '#4ECDC4', '#45B7D1', '#96CEB4'],
alpha=0.9, edgecolor='white', linewidth=2)
ax3.set_title('Statistiques de la base de données des embeddings', fontsize=14, color='white', fontweight='bold', pad=15)
ax3.set_ylabel('Count', fontsize=12, color='white')
ax3.tick_params(axis='x', labelsize=10, colors='white')
ax3.tick_params(axis='y', colors='white')
ax3.grid(True, alpha=0.3, color='#404040')
for bar, value in zip(bars, database_stats.values()):
ax3.text(bar.get_x() + bar.get_width()/2., bar.get_height() + max(database_stats.values()) * 0.01,
f'{value:,}', ha='center', va='bottom', color='white', fontweight='bold')
# Résumé des différents éléments du système
ax4 = plt.subplot(2, 2, 4)
ax4.set_facecolor('#2d2d2d')
ax4.axis('off')
if embeddings_ready > 0:
status = "Prêt pour la recherche de similarité"
elif recipes_with_images > 0:
status = "Prêt pour l'extraction d'embedding"
else:
status = "Aucune image trouvée"
combined_text = f"""
MODELE D'EXTRACTION D'EMBEDDING:
Architecture:
• Base: EfficientNetB0 (base pré-entraînée ImageNet)
• Output: {CONFIG['EMBEDDING_DIM_NATIVE']}-dim (raw)
• Traitement: GlobalAveragePooling2D + L2 normalize
Paramètres:
• Total: {total_params:,}
• Gelés: {total_params:,} (100% pré-entraînés)
• Entraînables: 0
Ce que fait ce modèle:
• Extraction d'embedding via EfficientNetB0 pré-entraîné
• GlobalAveragePooling2D → 1280 features
• L2 normalization
• Recherche par similarité Cosine
Ce que ça ne fait pas:
• Pas d'entraînement / transfer learning / fine-tuning
• Pas de custom Dense layers (tête custom)
• Pas d'optimisation spécifique aux recettes
Caractéristiques:
• Strategie: recherche de similarité Cosine
• Forces: reconnaissance correcte mais limité des images de recettes similaires
• Limites: ne comprend pas forcément toutes les subtilités propres aux recettes
• Vitesse: très rapide (pas d'entraînement), on extrait juste les embeddings
Statut: {status}
Dataset: {total_recipes:,} recettes, {recipes_with_images:,} avec images
"""
ax4.text(0.05, 0.95, combined_text, transform=ax4.transAxes, fontsize=9,
color='white', verticalalignment='top', fontfamily='monospace',
bbox=dict(boxstyle="round,pad=0.5", facecolor='#2d2d2d', alpha=0.8))
plt.tight_layout(pad=2.0, rect=[0, 0.02, 1, 0.92])
plt.show()
def visualize_embedding_space(self, sample_size=500):
""" Visualisation simplifiée des features EfficientNet """
if self.embeddings_db is None:
print("Il faut d'abord créer la base de données des embeddings!")
return
if len(self.embeddings_db) > sample_size:
indices = np.random.choice(len(self.embeddings_db), sample_size, replace=False)
sample_embeddings = self.embeddings_db[indices]
else:
sample_embeddings = self.embeddings_db
indices = np.arange(len(sample_embeddings))
# Création des plots
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(16, 6))
fig.patch.set_facecolor('#1a1a1a')
fig.suptitle('Analyse des features EfficientNet - Modèle Raw',
fontsize=18, color='white', fontweight='bold', y=0.95)
# Plot 1: Distribution des features
ax1.hist(sample_embeddings.flatten(), bins=50, alpha=0.7, color='skyblue', edgecolor='black')
ax1.set_facecolor('#2d2d2d')
ax1.set_title('Distribution des features (1280D)', fontsize=14, color='white', fontweight='bold')
ax1.set_xlabel('Valeurs des features', color='white')
ax1.set_ylabel('Fréquence', color='white')
ax1.tick_params(colors='white')
ax1.grid(True, alpha=0.3, color='#404040')
# Plot 2: Clustering visuel avec PCA
from sklearn.decomposition import PCA
pca = PCA(n_components=2)
embeddings_2d = pca.fit_transform(sample_embeddings)
# Analyser les propriétés visuelles basiques
image_properties = []
for i in indices:
try:
img = Image.open(self.image_paths[i])
img_array = np.array(img)
brightness = np.mean(img_array) / 255.0
image_properties.append(brightness)
except:
image_properties.append(0.5)
# Colorier par brightness
scatter = ax2.scatter(embeddings_2d[:, 0], embeddings_2d[:, 1],
c=image_properties, cmap='viridis', s=50, alpha=0.7)
ax2.set_facecolor('#2d2d2d')
ax2.set_title('Clusters visuels (PCA)', fontsize=14, color='white', fontweight='bold')
ax2.set_xlabel(f'PC1 ({pca.explained_variance_ratio_[0]:.1%})', color='white')
ax2.set_ylabel(f'PC2 ({pca.explained_variance_ratio_[1]:.1%})', color='white')
ax2.tick_params(colors='white')
ax2.grid(True, alpha=0.3, color='#404040')
plt.colorbar(scatter, ax=ax2, label='Luminosité')
plt.tight_layout()
plt.show()
# Statistiques simples
print(f"\nStatistiques:")
print(f"Échantillons analysés: {len(sample_embeddings)}")
print(f"Dimensions: {sample_embeddings.shape[1]} features EfficientNet")
print(f"Variance expliquée: {sum(pca.explained_variance_ratio_):.1%}")
print(f"Luminosité moyenne: {np.mean(image_properties):.2f}")
print("\nNote: Features génériques ImageNet sans apprentissage spécifique aux recettes")
def visualize_similarity_matrix(self, sample_size=20):
"""Visualisation de la matrice de similarité entre les recettes"""
if self.embeddings_db is None:
print("Il faut d'abord créer la base de données des embeddings!")
return
# Test sur des recettes aléatoires
indices = np.random.choice(len(self.embeddings_db), min(sample_size, len(self.embeddings_db)), replace=False)
sample_embeddings = self.embeddings_db[indices]
sample_titles = [self.image_to_recipe_map[self.image_paths[i]]['title'][:20] + "..."
for i in indices]
# Calculate similarity matrix
similarity_matrix = cosine_similarity(sample_embeddings)
# Heatmap
fig, ax = plt.subplots(figsize=(15, 12))
fig.patch.set_facecolor('#1a1a1a')
im = ax.imshow(similarity_matrix, cmap='viridis', aspect='auto')
cbar = plt.colorbar(im, ax=ax)
cbar.ax.tick_params(colors='white')
cbar.set_label('Similarité Cosine', color='white', fontweight='bold')
ax.set_xticks(range(len(sample_titles)))
ax.set_yticks(range(len(sample_titles)))
ax.set_xticklabels(sample_titles, rotation=45, ha='right', color='white', fontsize=9)
ax.set_yticklabels(sample_titles, color='white', fontsize=9)
# Add text annotations
for i in range(len(sample_titles)):
for j in range(len(sample_titles)):
text = ax.text(j, i, f'{similarity_matrix[i, j]:.2f}',
ha="center", va="center", color="white", fontsize=8, fontweight='bold')
ax.set_title('MATRICE DE SIMILARITE',
fontsize=16, color='white', fontweight='bold', pad=20)
ax.set_facecolor('#2d2d2d')
plt.tight_layout()
plt.show()
print(f"Matrice de similarité pour {len(sample_titles)} recettes")
# Core functionality (unchanged)
def search_similar_recipes(self, query_image_path, top_k=3):
"""Rechercher le top-k des recettes similaires à une image de recette"""
if self.embeddings_db is None:
print("Il faut d'abord créer la base de données des embeddings!")
return None
print(f"Recherche de {top_k} recettes similaires...")
try:
img = tf.keras.preprocessing.image.load_img(
query_image_path, target_size=(CONFIG['IMG_SIZE'], CONFIG['IMG_SIZE'])
)
img = tf.keras.preprocessing.image.img_to_array(img)
img = tf.keras.applications.efficientnet.preprocess_input(img)
img = np.expand_dims(img, axis=0)
query_embedding = self.model.predict(img, verbose=0)
except Exception as e:
print(f"Erreur chargement de l'image de recette: {e}")
return None
# Calcul des similarités
similarities = cosine_similarity(query_embedding, self.embeddings_db)[0]
# Get top k
top_indices = np.argsort(similarities)[::-1][:top_k]
results = []
for rank, idx in enumerate(top_indices):
img_path = self.image_paths[idx]
similarity_score = similarities[idx]
if img_path in self.image_to_recipe_map:
recipe_info = self.image_to_recipe_map[img_path].copy()
recipe_info['rank'] = rank + 1
recipe_info['similarity'] = similarity_score
recipe_info['similar_image_path'] = img_path
results.append(recipe_info)
print(f"{len(results)} recipes similaires trouvées")
return results
def display_results(self, query_image_path, results):
"""Afficher les résultats de la recherche"""
if not results:
print("Aucun résultat")
return
fig, axes = plt.subplots(1, len(results) + 1, figsize=(20, 6))
fig.patch.set_facecolor('#1a1a1a')
fig.suptitle('IMAGE → TOP 3 RECETTES SIMILAIRES',
fontsize=18, color='white', fontweight='bold', y=0.95)
colors = ['#FF6B9D', '#4ECDC4', '#45B7D1', '#96CEB4']
# Image en input
try:
query_img = Image.open(query_image_path)
axes[0].imshow(query_img)
axes[0].set_title("INPUT", fontsize=14, color=colors[0], fontweight='bold', pad=15)
axes[0].axis('off')
# Add border
for spine in axes[0].spines.values():
spine.set_visible(True)
spine.set_color(colors[0])
spine.set_linewidth(3)
except:
axes[0].text(0.5, 0.5, "Pas d\'input", ha='center', va='center',
transform=axes[0].transAxes, color='white', fontsize=12)
axes[0].set_title("INPUT", fontsize=14, color=colors[0], fontweight='bold')
axes[0].axis('off')
# Similar images
for i, result in enumerate(results):
try:
result_img = Image.open(result['similar_image_path'])
axes[i+1].imshow(result_img)
title = result['title'][:25] + "..." if len(result['title']) > 25 else result['title']
similarity = result['similarity']
axes[i+1].set_title(f"#{result['rank']}: {title}\n Similarity: {similarity:.3f}",
fontsize=12, color=colors[i+1], fontweight='bold', pad=15)
axes[i+1].axis('off')
# Add colored border
for spine in axes[i+1].spines.values():
spine.set_visible(True)
spine.set_color(colors[i+1])
spine.set_linewidth(3)
except:
axes[i+1].text(0.5, 0.5, f"Image #{result['rank']}\nabsente",
ha='center', va='center', transform=axes[i+1].transAxes,
color='white', fontsize=11)
axes[i+1].set_title(f"#{result['rank']}: ERREUR", fontsize=12, color='red', fontweight='bold')
axes[i+1].axis('off')
plt.tight_layout()
plt.show()
print("\n" + "="*80)
print("RECETTES SIMILAIRES")
print("="*80)
for result in results:
print(f"\nRANG #{result['rank']}: {result['similarity']:.4f}")
print(f"{result['title']}")
print("-" * 60)
# Ingredients
ingredients = str(result['ingredients'])
if len(ingredients) > 300:
ingredients = ingredients[:300] + "..."
print(f"Ingredients:\n{ingredients}")
print("-" * 60)
# Instructions
instructions = str(result['instructions'])
if len(instructions) > 400:
instructions = instructions[:400] + "..."
print(f"Instructions:\n{instructions}")
print("=" * 80)
def test_with_random_image(self, top_k=3):
"""Test avec une image aléatoire de la base de données"""
if not self.image_paths:
print("Aucune image disponible")
return
random_idx = np.random.randint(0, len(self.image_paths))
test_image_path = self.image_paths[random_idx]
print(f"Test avec une image aléatoire: {test_image_path}")
results = self.search_similar_recipes(test_image_path, top_k)
if results:
self.display_results(test_image_path, results)
else:
print("Aucun résultat")
print("Système créé!")
print("Cette classe permet d'extraire les embeddings des images de recettes et de les rechercher par similarité")
print("\nMéthodes disponibles:")
print("- retrieval_system.visualize_model_architecture()")
print("- retrieval_system.visualize_embedding_space()")
print("- retrieval_system.visualize_similarity_matrix()")
Système créé! Cette classe permet d'extraire les embeddings des images de recettes et de les rechercher par similarité Méthodes disponibles: - retrieval_system.visualize_model_architecture() - retrieval_system.visualize_embedding_space() - retrieval_system.visualize_similarity_matrix()
8. Extraction des embeddings depuis le dataset¶
In [9]:
# Extraction des features
print("Initialisation du système...")
# Check if recipes_with_images exists
if 'recipes_with_images' in globals() and len(recipes_with_images) > 0:
print(f"{len(recipes_with_images)} recettes avec images")
retrieval_system = RecipeImageRetrieval(image_retrieval_model, recipes_with_images)
retrieval_system.prepare_database()
print("Système initialisé avec succès!")
print(f"La base de données contient {len(retrieval_system.image_paths)} images")
sample_titles = [retrieval_system.image_to_recipe_map[path]['title']
for path in list(retrieval_system.image_paths)[:5]]
print(f"Exemples de recettes: {sample_titles}")
else:
print("'recipes_with_images' n'est pas défini!")
print("Veuillez d'abord exécuter les cellules précédentes pour charger et traiter les données.")
# Extraction des embeddings
print("Extraction des embeddings...")
retrieval_system.build_embeddings_database(batch_size=16)
print("\nProchaines étapes:")
print("1. Test sur une image aléatoire du dataset: retrieval_system.test_with_random_image()")
print("2. Test sur une image de recette: retrieval_system.search_similar_recipes('path/to/image.jpg')")
Initialisation du système...
13463 recettes avec images
On prépare la base de données des images de recipes_with_images...
13463 images trouvées
Mapping image-to-recipe créé!
Système initialisé avec succès!
La base de données contient 13463 images
Exemples de recettes: ['Miso-Butter Roast Chicken With Acorn Squash Panzanella', 'Crispy Salt and Pepper Potatoes', 'Thanksgiving Mac and Cheese', 'Italian Sausage and Bread Stuffing', "Newton's Law"]
Extraction des embeddings...
Extraction des embeddings...
Batch 1/842...
Batch 2/842...
Batch 3/842...
Batch 4/842...
Batch 5/842...
Batch 6/842...
Batch 7/842...
Batch 8/842...
Batch 9/842...
Batch 10/842...
Batch 11/842...
Batch 12/842...
Batch 13/842...
Batch 14/842...
Batch 15/842...
Batch 16/842...
Batch 17/842...
Batch 18/842...
Batch 19/842...
Batch 20/842...
Batch 21/842...
Batch 22/842...
Batch 23/842...
Batch 24/842...
Batch 25/842...
Batch 26/842...
Batch 27/842...
Batch 28/842...
Batch 29/842...
Batch 30/842...
Batch 31/842...
Batch 32/842...
Batch 33/842...
Batch 34/842...
Batch 35/842...
Batch 36/842...
Batch 37/842...
Batch 38/842...
Batch 39/842...
Batch 40/842...
Batch 41/842...
Batch 42/842...
Batch 43/842...
Batch 44/842...
Batch 45/842...
Batch 46/842...
Batch 47/842...
Batch 48/842...
Batch 49/842...
Batch 50/842...
Batch 51/842...
Batch 52/842...
Batch 53/842...
Batch 54/842...
Batch 55/842...
Batch 56/842...
Batch 57/842...
Batch 58/842...
Batch 59/842...
Batch 60/842...
Batch 61/842...
Batch 62/842...
Batch 63/842...
Batch 64/842...
Batch 65/842...
Batch 66/842...
Batch 67/842...
Batch 68/842...
Batch 69/842...
Batch 70/842...
Batch 71/842...
Batch 72/842...
Batch 73/842...
Batch 74/842...
Batch 75/842...
Batch 76/842...
Batch 77/842...
Batch 78/842...
Batch 79/842...
Batch 80/842...
Batch 81/842...
Batch 82/842...
Batch 83/842...
Batch 84/842...
Batch 85/842...
Batch 86/842...
Batch 87/842...
Batch 88/842...
Batch 89/842...
Batch 90/842...
Batch 91/842...
Batch 92/842...
Batch 93/842...
Batch 94/842...
Batch 95/842...
Batch 96/842...
Batch 97/842...
Batch 98/842...
Batch 99/842...
Batch 100/842...
Batch 101/842...
Batch 102/842...
Batch 103/842...
Batch 104/842...
Batch 105/842...
Batch 106/842...
Batch 107/842...
Batch 108/842...
Batch 109/842...
Batch 110/842...
Batch 111/842...
Batch 112/842...
Batch 113/842...
Batch 114/842...
Batch 115/842...
Batch 116/842...
Batch 117/842...
Batch 118/842...
Batch 119/842...
Batch 120/842...
Batch 121/842...
Batch 122/842...
Batch 123/842...
Batch 124/842...
Batch 125/842...
Batch 126/842...
Batch 127/842...
Batch 128/842...
Batch 129/842...
Batch 130/842...
Batch 131/842...
Batch 132/842...
Batch 133/842...
Batch 134/842...
Batch 135/842...
Batch 136/842...
Batch 137/842...
Batch 138/842...
Batch 139/842...
Batch 140/842...
Batch 141/842...
Batch 142/842...
Batch 143/842...
Batch 144/842...
Batch 145/842...
Batch 146/842...
Batch 147/842...
Batch 148/842...
Batch 149/842...
Batch 150/842...
Batch 151/842...
Batch 152/842...
Batch 153/842...
Batch 154/842...
Batch 155/842...
Batch 156/842...
Batch 157/842...
Batch 158/842...
Batch 159/842...
Batch 160/842...
Batch 161/842...
Batch 162/842...
Batch 163/842...
Batch 164/842...
Batch 165/842...
Batch 166/842...
Batch 167/842...
Batch 168/842...
Batch 169/842...
Batch 170/842...
Batch 171/842...
Batch 172/842...
Batch 173/842...
Batch 174/842...
Batch 175/842...
Batch 176/842...
Batch 177/842...
Batch 178/842...
Batch 179/842...
Batch 180/842...
Batch 181/842...
Batch 182/842...
Batch 183/842...
Batch 184/842...
Batch 185/842...
Batch 186/842...
Batch 187/842...
Batch 188/842...
Batch 189/842...
Batch 190/842...
Batch 191/842...
Batch 192/842...
Batch 193/842...
Batch 194/842...
Batch 195/842...
Batch 196/842...
Batch 197/842...
Batch 198/842...
Batch 199/842...
Batch 200/842...
Batch 201/842...
Batch 202/842...
Batch 203/842...
Batch 204/842...
Batch 205/842...
Batch 206/842...
Batch 207/842...
Batch 208/842...
Batch 209/842...
Batch 210/842...
Batch 211/842...
Batch 212/842...
Batch 213/842...
Batch 214/842...
Batch 215/842...
Batch 216/842...
Batch 217/842...
Batch 218/842...
Batch 219/842...
Batch 220/842...
Batch 221/842...
Batch 222/842...
Batch 223/842...
Batch 224/842...
Batch 225/842...
Batch 226/842...
Batch 227/842...
Batch 228/842...
Batch 229/842...
Batch 230/842...
Batch 231/842...
Batch 232/842...
Batch 233/842...
Batch 234/842...
Batch 235/842...
Batch 236/842...
Batch 237/842...
Batch 238/842...
Batch 239/842...
Batch 240/842...
Batch 241/842...
Batch 242/842...
Batch 243/842...
Batch 244/842...
Batch 245/842...
Batch 246/842...
Batch 247/842...
Batch 248/842...
Batch 249/842...
Batch 250/842...
Batch 251/842...
Batch 252/842...
Batch 253/842...
Batch 254/842...
Batch 255/842...
Batch 256/842...
Batch 257/842...
Batch 258/842...
Batch 259/842...
Batch 260/842...
Batch 261/842...
Batch 262/842...
Batch 263/842...
Batch 264/842...
Batch 265/842...
Batch 266/842...
Batch 267/842...
Batch 268/842...
Batch 269/842...
Batch 270/842...
Batch 271/842...
Batch 272/842...
Batch 273/842...
Batch 274/842...
Batch 275/842...
Batch 276/842...
Batch 277/842...
Batch 278/842...
Batch 279/842...
Batch 280/842...
Batch 281/842...
Batch 282/842...
Batch 283/842...
Batch 284/842...
Batch 285/842...
Batch 286/842...
Batch 287/842...
Batch 288/842...
Batch 289/842...
Batch 290/842...
Batch 291/842...
Batch 292/842...
Batch 293/842...
Batch 294/842...
Batch 295/842...
Batch 296/842...
Batch 297/842...
Batch 298/842...
Batch 299/842...
Batch 300/842...
Batch 301/842...
Batch 302/842...
Batch 303/842...
Batch 304/842...
Batch 305/842...
Batch 306/842...
Batch 307/842...
Batch 308/842...
Batch 309/842...
Batch 310/842...
Batch 311/842...
Batch 312/842...
Batch 313/842...
Batch 314/842...
Batch 315/842...
Batch 316/842...
Batch 317/842...
Batch 318/842...
Batch 319/842...
Batch 320/842...
Batch 321/842...
Batch 322/842...
Batch 323/842...
Batch 324/842...
Batch 325/842...
Batch 326/842...
Batch 327/842...
Batch 328/842...
Batch 329/842...
Batch 330/842...
Batch 331/842...
Batch 332/842...
Batch 333/842...
Batch 334/842...
Batch 335/842...
Batch 336/842...
Batch 337/842...
Batch 338/842...
Batch 339/842...
Batch 340/842...
Batch 341/842...
Batch 342/842...
Batch 343/842...
Batch 344/842...
Batch 345/842...
Batch 346/842...
Batch 347/842...
Batch 348/842...
Batch 349/842...
Batch 350/842...
Batch 351/842...
Batch 352/842...
Batch 353/842...
Batch 354/842...
Batch 355/842...
Batch 356/842...
Batch 357/842...
Batch 358/842...
Batch 359/842...
Batch 360/842...
Batch 361/842...
Batch 362/842...
Batch 363/842...
Batch 364/842...
Batch 365/842...
Batch 366/842...
Batch 367/842...
Batch 368/842...
Batch 369/842...
Batch 370/842...
Batch 371/842...
Batch 372/842...
Batch 373/842...
Batch 374/842...
Batch 375/842...
Batch 376/842...
Batch 377/842...
Batch 378/842...
Batch 379/842...
Batch 380/842...
Batch 381/842...
Batch 382/842...
Batch 383/842...
Batch 384/842...
Batch 385/842...
Batch 386/842...
Batch 387/842...
Batch 388/842...
Batch 389/842...
Batch 390/842...
Batch 391/842...
Batch 392/842...
Batch 393/842...
Batch 394/842...
Batch 395/842...
Batch 396/842...
Batch 397/842...
Batch 398/842...
Batch 399/842...
Batch 400/842...
Batch 401/842...
Batch 402/842...
Batch 403/842...
Batch 404/842...
Batch 405/842...
Batch 406/842...
Batch 407/842...
Batch 408/842...
Batch 409/842...
Batch 410/842...
Batch 411/842...
Batch 412/842...
Batch 413/842...
Batch 414/842...
Batch 415/842...
Batch 416/842...
Batch 417/842...
Batch 418/842...
Batch 419/842...
Batch 420/842...
Batch 421/842...
Batch 422/842...
Batch 423/842...
Batch 424/842...
Batch 425/842...
Batch 426/842...
Batch 427/842...
Batch 428/842...
Batch 429/842...
Batch 430/842...
Batch 431/842...
Batch 432/842...
Batch 433/842...
Batch 434/842...
Batch 435/842...
Batch 436/842...
Batch 437/842...
Batch 438/842...
Batch 439/842...
Batch 440/842...
Batch 441/842...
Batch 442/842...
Batch 443/842...
Batch 444/842...
Batch 445/842...
Batch 446/842...
Batch 447/842...
Batch 448/842...
Batch 449/842...
Batch 450/842...
Batch 451/842...
Batch 452/842...
Batch 453/842...
Batch 454/842...
Batch 455/842...
Batch 456/842...
Batch 457/842...
Batch 458/842...
Batch 459/842...
Batch 460/842...
Batch 461/842...
Batch 462/842...
Batch 463/842...
Batch 464/842...
Batch 465/842...
Batch 466/842...
Batch 467/842...
Batch 468/842...
Batch 469/842...
Batch 470/842...
Batch 471/842...
Batch 472/842...
Batch 473/842...
Batch 474/842...
Batch 475/842...
Batch 476/842...
Batch 477/842...
Batch 478/842...
Batch 479/842...
Batch 480/842...
Batch 481/842...
Batch 482/842...
Batch 483/842...
Batch 484/842...
Batch 485/842...
Batch 486/842...
Batch 487/842...
Batch 488/842...
Batch 489/842...
Batch 490/842...
Batch 491/842...
Batch 492/842...
Batch 493/842...
Batch 494/842...
Batch 495/842...
Batch 496/842...
Batch 497/842...
Batch 498/842...
Batch 499/842...
Batch 500/842...
Batch 501/842...
Batch 502/842...
Batch 503/842...
Batch 504/842...
Batch 505/842...
Batch 506/842...
Batch 507/842...
Batch 508/842...
Batch 509/842...
Batch 510/842...
Batch 511/842...
Batch 512/842...
Batch 513/842...
Batch 514/842...
Batch 515/842...
Batch 516/842...
Batch 517/842...
Batch 518/842...
Batch 519/842...
Batch 520/842...
Batch 521/842...
Batch 522/842...
Batch 523/842...
Batch 524/842...
Batch 525/842...
Batch 526/842...
Batch 527/842...
Batch 528/842...
Batch 529/842...
Batch 530/842...
Batch 531/842...
Batch 532/842...
Batch 533/842...
Batch 534/842...
Batch 535/842...
Batch 536/842...
Batch 537/842...
Batch 538/842...
Batch 539/842...
Batch 540/842...
Batch 541/842...
Batch 542/842...
Batch 543/842...
Batch 544/842...
Batch 545/842...
Batch 546/842...
Batch 547/842...
Batch 548/842...
Batch 549/842...
Batch 550/842...
Batch 551/842...
Batch 552/842...
Batch 553/842...
Batch 554/842...
Batch 555/842...
Batch 556/842...
Batch 557/842...
Batch 558/842...
Batch 559/842...
Batch 560/842...
Batch 561/842...
Batch 562/842...
Batch 563/842...
Batch 564/842...
Batch 565/842...
Batch 566/842...
Batch 567/842...
Batch 568/842...
Batch 569/842...
Batch 570/842...
Batch 571/842...
Batch 572/842...
Batch 573/842...
Batch 574/842...
Batch 575/842...
Batch 576/842...
Batch 577/842...
Batch 578/842...
Batch 579/842...
Batch 580/842...
Batch 581/842...
Batch 582/842...
Batch 583/842...
Batch 584/842...
Batch 585/842...
Batch 586/842...
Batch 587/842...
Batch 588/842...
Batch 589/842...
Batch 590/842...
Batch 591/842...
Batch 592/842...
Batch 593/842...
Batch 594/842...
Batch 595/842...
Batch 596/842...
Batch 597/842...
Batch 598/842...
Batch 599/842...
Batch 600/842...
Batch 601/842...
Batch 602/842...
Batch 603/842...
Batch 604/842...
Batch 605/842...
Batch 606/842...
Batch 607/842...
Batch 608/842...
Batch 609/842...
Batch 610/842...
Batch 611/842...
Batch 612/842...
Batch 613/842...
Batch 614/842...
Batch 615/842...
Batch 616/842...
Batch 617/842...
Batch 618/842...
Batch 619/842...
Batch 620/842...
Batch 621/842...
Batch 622/842...
Batch 623/842...
Batch 624/842...
Batch 625/842...
Batch 626/842...
Batch 627/842...
Batch 628/842...
Batch 629/842...
Batch 630/842...
Batch 631/842...
Batch 632/842...
Batch 633/842...
Batch 634/842...
Batch 635/842...
Batch 636/842...
Batch 637/842...
Batch 638/842...
Batch 639/842...
Batch 640/842...
Batch 641/842...
Batch 642/842...
Batch 643/842...
Batch 644/842...
Batch 645/842...
Batch 646/842...
Batch 647/842...
Batch 648/842...
Batch 649/842...
Batch 650/842...
Batch 651/842...
Batch 652/842...
Batch 653/842...
Batch 654/842...
Batch 655/842...
Batch 656/842...
Batch 657/842...
Batch 658/842...
Batch 659/842...
Batch 660/842...
Batch 661/842...
Batch 662/842...
Batch 663/842...
Batch 664/842...
Batch 665/842...
Batch 666/842...
Batch 667/842...
Batch 668/842...
Batch 669/842...
Batch 670/842...
Batch 671/842...
Batch 672/842...
Batch 673/842...
Batch 674/842...
Batch 675/842...
Batch 676/842...
Batch 677/842...
Batch 678/842...
Batch 679/842...
Batch 680/842...
Batch 681/842...
Batch 682/842...
Batch 683/842...
Batch 684/842...
Batch 685/842...
Batch 686/842...
Batch 687/842...
Batch 688/842...
Batch 689/842...
Batch 690/842...
Batch 691/842...
Batch 692/842...
Batch 693/842...
Batch 694/842...
Batch 695/842...
Batch 696/842...
Batch 697/842...
Batch 698/842...
Batch 699/842...
Batch 700/842...
Batch 701/842...
Batch 702/842...
Batch 703/842...
Batch 704/842...
Batch 705/842...
Batch 706/842...
Batch 707/842...
Batch 708/842...
Batch 709/842...
Batch 710/842...
Batch 711/842...
Batch 712/842...
Batch 713/842...
Batch 714/842...
Batch 715/842...
Batch 716/842...
Batch 717/842...
Batch 718/842...
Batch 719/842...
Batch 720/842...
Batch 721/842...
Batch 722/842...
Batch 723/842...
Batch 724/842...
Batch 725/842...
Batch 726/842...
Batch 727/842...
Batch 728/842...
Batch 729/842...
Batch 730/842...
Batch 731/842...
Batch 732/842...
Batch 733/842...
Batch 734/842...
Batch 735/842...
Batch 736/842...
Batch 737/842...
Batch 738/842...
Batch 739/842...
Batch 740/842...
Batch 741/842...
Batch 742/842...
Batch 743/842...
Batch 744/842...
Batch 745/842...
Batch 746/842...
Batch 747/842...
Batch 748/842...
Batch 749/842...
Batch 750/842...
Batch 751/842...
Batch 752/842...
Batch 753/842...
Batch 754/842...
Batch 755/842...
Batch 756/842...
Batch 757/842...
Batch 758/842...
Batch 759/842...
Batch 760/842...
Batch 761/842...
Batch 762/842...
Batch 763/842...
Batch 764/842...
Batch 765/842...
Batch 766/842...
Batch 767/842...
Batch 768/842...
Batch 769/842...
Batch 770/842...
Batch 771/842...
Batch 772/842...
Batch 773/842...
Batch 774/842...
Batch 775/842...
Batch 776/842...
Batch 777/842...
Batch 778/842...
Batch 779/842...
Batch 780/842...
Batch 781/842...
Batch 782/842...
Batch 783/842...
Batch 784/842...
Batch 785/842...
Batch 786/842...
Batch 787/842...
Batch 788/842...
Batch 789/842...
Batch 790/842...
Batch 791/842...
Batch 792/842...
Batch 793/842...
Batch 794/842...
Batch 795/842...
Batch 796/842...
Batch 797/842...
Batch 798/842...
Batch 799/842...
Batch 800/842...
Batch 801/842...
Batch 802/842...
Batch 803/842...
Batch 804/842...
Batch 805/842...
Batch 806/842...
Batch 807/842...
Batch 808/842...
Batch 809/842...
Batch 810/842...
Batch 811/842...
Batch 812/842...
Batch 813/842...
Batch 814/842...
Batch 815/842...
Batch 816/842...
Batch 817/842...
Batch 818/842...
Batch 819/842...
Batch 820/842...
Batch 821/842...
Batch 822/842...
Batch 823/842...
Batch 824/842...
Batch 825/842...
Batch 826/842...
Batch 827/842...
Batch 828/842...
Batch 829/842...
Batch 830/842...
Batch 831/842...
Batch 832/842...
Batch 833/842...
Batch 834/842...
Batch 835/842...
Batch 836/842...
Batch 837/842...
Batch 838/842...
Batch 839/842...
Batch 840/842...
Batch 841/842...
Batch 842/842...
Base de données des embeddings créée: 13463 embeddings
Shape embeddings: (13463, 1280)
Prochaines étapes:
1. Test sur une image aléatoire du dataset: retrieval_system.test_with_random_image()
2. Test sur une image de recette: retrieval_system.search_similar_recipes('path/to/image.jpg')
9. Visualisations des caractéristiques du modèle et des embeddings extraits¶
In [10]:
retrieval_system.visualize_embedding_space()
retrieval_system.visualize_similarity_matrix()
retrieval_system.visualize_model_architecture()
Statistiques: Échantillons analysés: 500 Dimensions: 1280 features EfficientNet Variance expliquée: 11.1% Luminosité moyenne: 0.56 Note: Features génériques ImageNet sans apprentissage spécifique aux recettes
Matrice de similarité pour 20 recettes
10. Sauvegarde complète du système¶
In [11]:
import pickle
def save_complete_system(retrieval_system):
"""Save the complete retrieval system"""
print(" Saving complete retrieval system...")
# Sauvegarde du modèle
retrieval_system.model.save("./raw/embeddings_recipe_image_retrieval_model_raw.keras")
print("Modèle sauvegardé: ./raw/embeddings_recipe_image_retrieval_model_raw.keras")
# Sauvegarde de la base de données des embeddings
database_path = "./raw/recipe_embeddings_database_raw"
metadata_path = "./raw/recipe_embeddings_database_metadata_raw"
np.save(f"{database_path}.npy", retrieval_system.embeddings_db)
print(f"Embeddings sauvegardés: {database_path}.npy ({retrieval_system.embeddings_db.shape})")
# Sauvegarde des métadonnées (en utilisant les attributs actuels)
metadata = {
'image_paths': retrieval_system.image_paths,
'image_to_recipe_map': retrieval_system.image_to_recipe_map
}
with open(f"{metadata_path}.pkl", 'wb') as f:
pickle.dump(metadata, f)
print(f"Metadata sauvegardées: ./raw/recipe_embeddings_database_metadata_raw.pkl")
# Sauvegarde du DataFrame 'recipes_with_images'
recipes_df_path = "./data/recipes_with_images_dataframe.pkl"
retrieval_system.recipes_df.to_pickle(recipes_df_path)
print(f"DataFrame 'recipes_with_images' sauvegardé: {recipes_df_path}")
model_size = os.path.getsize("./raw/embeddings_recipe_image_retrieval_model_raw.keras") / (1024**2)
embeddings_size = os.path.getsize(f"{database_path}.npy") / (1024**2)
metadata_size = os.path.getsize(f"{metadata_path}.pkl") / (1024**2)
df_size = os.path.getsize(recipes_df_path) / (1024**2)
print(f"\nTaille des fichiers:")
print(f"Model: {model_size:.1f} MB")
print(f"Embeddings: {embeddings_size:.1f} MB")
print(f"Metadata: {metadata_size:.1f} MB")
print(f" DataFrame: {df_size:.1f} MB")
print(f"\nSauvegarde complète du modèle et des {len(retrieval_system.embeddings_db):,} embeddings effectuée!")
save_complete_system(retrieval_system)
Saving complete retrieval system...
Modèle sauvegardé: ./raw/embeddings_recipe_image_retrieval_model_raw.keras
Embeddings sauvegardés: ./raw/recipe_embeddings_database_raw.npy ((13463, 1280))
Metadata sauvegardées: ./raw/recipe_embeddings_database_metadata_raw.pkl
DataFrame 'recipes_with_images' sauvegardé: ./data/recipes_with_images_dataframe.pkl
Taille des fichiers:
Model: 16.2 MB
Embeddings: 65.7 MB
Metadata: 22.2 MB
DataFrame: 73.2 MB
Sauvegarde complète du modèle et des 13,463 embeddings effectuée!
11. Test du recherche avec une image aléatoire du dataset¶
In [12]:
# Test avec une image aléatoire du dataset
# Si possible, on recharge le système depuis les fichiers sauvegardés => parfait pour préparer le déploiement
import tensorflow as tf
import pickle
import numpy as np
import os
import pandas as pd
def l2_norm(x):
return tf.nn.l2_normalize(x, axis=1)
def same_shape(x):
return x
CONFIG_STANDALONE = {
'IMG_SIZE': 224,
}
def ensure_retrieval_system():
"""Init du système de recherche avec chargement depuis les fichiers sauvegardés si possible"""
global retrieval_system
# On vérifie si le système existe déjà et est prêt
if 'retrieval_system' in globals() and retrieval_system is not None and retrieval_system.embeddings_db is not None:
print("Utilisation du système existant en mémoire")
return True
print("Chargement du système depuis les fichiers sauvegardés...")
try:
# On a besoin des custom l2_norm et same_shape pour le chargement du modèle
custom_objects = {
'l2_norm': l2_norm,
'same_shape': same_shape,
}
# Chemin du modèle sauvegardé précédemment
model_paths = [
"./raw/embeddings_recipe_image_retrieval_model_raw.keras"
]
model = None
for model_path in model_paths:
if os.path.exists(model_path):
try:
model = tf.keras.models.load_model(model_path,
compile=False,
safe_mode=False,
custom_objects=custom_objects)
print(f"Modèle chargé depuis {model_path}")
break
except Exception as e:
print(f"Echec de chargement depuis {model_path}: {e}")
continue
if model is None:
print(" Could not load model from any location")
return False
# Chargement de la base de données des embeddings
embeddings_db = np.load("./raw/recipe_embeddings_database_raw.npy")
print(f"Embeddings chargés: {embeddings_db.shape}")
# Chargement des métadonnées
with open("./raw/recipe_embeddings_database_metadata_raw.pkl", 'rb') as f:
metadata = pickle.load(f)
print("Métadonnées chargées")
# Chargement du DataFrame 'recipes_with_images'
recipes_df_path = "./data/recipes_with_images_dataframe.pkl"
if os.path.exists(recipes_df_path):
recipes_with_images = pd.read_pickle(recipes_df_path)
print(f"DataFrame chargée: {len(recipes_with_images)} recipes")
else:
print("Fichier recipes_with_images_dataframe.pkl introuvable!")
print("Il faut d'abord sauvegarder le DataFrame 'recipes_with_images' dans les cellules précédentes")
return False
# On crée le système avec les différents éléments chargés
retrieval_system = RecipeImageRetrieval(model, recipes_with_images)
retrieval_system.embeddings_db = embeddings_db
retrieval_system.image_paths = metadata['image_paths']
retrieval_system.image_to_recipe_map = metadata['image_to_recipe_map']
print(f"Système chargé avec {len(embeddings_db):,} embeddings!")
return True
except FileNotFoundError as e:
print(f"Fichier introuvable: {e}")
print("Vérifiez que ces fichiers existent:")
print("- ./raw/embeddings_recipe_image_retrieval_model_raw.keras")
print("- ./raw/recipe_embeddings_database_raw.npy")
print("- ./raw/recipe_embeddings_database_metadata_raw.pkl")
print("- ./data/recipes_with_images_dataframe_raw.pkl")
return False
except Exception as e:
print(f"Erreur lors du chargement du système: {e}")
import traceback
traceback.print_exc()
return False
# Main testing logic
if ensure_retrieval_system():
print("\nTest avec une image aléatoire du dataset...")
retrieval_system.test_with_random_image(top_k=3)
else:
print("\nImpossible d'initialiser le système")
Utilisation du système existant en mémoire Test avec une image aléatoire du dataset... Test avec une image aléatoire: C:\Users\Naturel\.cache\kagglehub\datasets\pes12017000148\food-ingredients-and-recipe-dataset-with-images\versions\1\Food Images\Food Images\arugula-and-ricotta-calzones-233791.jpg Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 1.0000 Arugula and Ricotta Calzones ------------------------------------------------------------ Ingredients: ['1 large garlic clove, minced', '2 tablespoons extra-virgin olive oil', '5 oz baby arugula (8 cups packed)', '6 oz whole-milk ricotta (2/3 cup)', '3 oz whole-milk mozzarella, coarsely grated', '2 tablespoonsfinely grated Parmigiano-Reggiano', '1 large egg yolk', '1/4 teaspoon salt', '1/8 teaspoon b... ------------------------------------------------------------ Instructions: Put oven rack in lower third of oven and preheat oven to 450°F. Cook garlic in oil in a 12-inch heavy skillet over moderate heat, stirring frequently, until golden, 1 to 2 minutes. Add arugula and cook, stirring frequently, until wilted, 2 to 3 minutes. Transfer to a sieve and press hard on arugula to squeeze out as much excess liquid as possible, then coarsely chop. Stir together ricotta, mozzare... ================================================================================ RANG #2: 0.6778 Lemon-Rhubarb Chicken ------------------------------------------------------------ Ingredients: ['5 tablespoons olive oil, divided', '2 tablespoons plus 1/4 cup chopped shallots', '4 1/2 cups diced rhubarb, divided', '1 tablespoon fresh lemon juice', '2 teaspoons finely grated lemon peel, divided', '1/4 cup (1/2 stick) butter', '1/2 cup sliced unpeeled fresh ginger', '3/4 cup sugar', '6 tables... ------------------------------------------------------------ Instructions: Heat 2 tablespoons olive oil in heavy large skillet over medium-high heat. Add 2 tablespoons chopped shallots and 2 cups rhubarb; sauté until soft, about 5 minutes. Stir in lemon juice and 1 teaspoon lemon peel. Season with salt and pepper. Cool rhubarb stuffing. Melt butter in heavy large saucepan over low heat. Add 21/2 cups rhubarb, 1/4 cup shallots, and ginger; sauté until soft, about 10 minut... ================================================================================ RANG #3: 0.6708 Shrimp and Cotija Enchiladas with Salsa Verde and Crema Mexicana ------------------------------------------------------------ Ingredients: ['4 tablespoons olive oil, divided', '2 1/2 pounds tomatillos,* husked, rinsed', '4 large fresh poblano chiles,* halved lengthwise, cored, seeded', '4 unpeeled garlic cloves', '2 cups (packed) coarsely chopped fresh cilantro plus additional for garnish', 'plus additional for garnish', '1 cup (packed... ------------------------------------------------------------ Instructions: Preheat broiler. Line large rimmed baking sheet with foil; brush lightly with 1 tablespoon olive oil. Arrange tomatillos, poblano chiles, cut side down, and garlic cloves on prepared baking sheet. Broil until tomatillos and chiles begin to soften and blacken in several spots, watching closely to prevent burning, about 10 minutes (do not turn). Remove from oven. Let stand until cool enough to handl... ================================================================================
12. Recherches sur des images inconnues du système¶
In [13]:
import tensorflow as tf
import numpy as np
if 'retrieval_system' not in globals() or retrieval_system.embeddings_db is None:
print("Chargement du système sauvegardé...")
try:
# Chargement du modèle
model = tf.keras.models.load_model("./raw/recipe_image_retrieval_model_raw.keras")
embeddings_db = np.load("./raw/recipe_embeddings_database_raw.npy")
import pickle
with open("./raw/recipe_embeddings_database_metadata_raw.pkl", 'rb') as f:
metadata = pickle.load(f)
# On crée le système avec les différents éléments chargés
retrieval_system = RecipeImageRetrieval(model, recipes_with_images)
retrieval_system.embeddings_db = embeddings_db
retrieval_system.image_paths = metadata['image_paths']
retrieval_system.image_to_recipe_map = metadata['image_to_recipe_map']
except FileNotFoundError:
print("Aucun système sauvegardé trouvé.")
retrieval_system = None
except Exception as e:
print(f"Erreur lors du chargement: {e}")
retrieval_system = None
if retrieval_system and retrieval_system.embeddings_db is not None:
# Test sur les images 1.jpg à 8.jpg
import os
for i in range(1, 9):
test_image = f"./test_recipes/{i}.jpg"
if os.path.exists(test_image):
print(f"\n{'='*50}")
print(f"TEST IMAGE {i}.jpg")
print(f"{'='*50}")
results = retrieval_system.search_similar_recipes(test_image, top_k=3)
if results:
retrieval_system.display_results(test_image, results)
else:
print("Aucun résultat trouvé")
else:
print(f"Image {i}.jpg introuvable")
else:
print("Système non disponible")
================================================== TEST IMAGE 1.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.6437 Lamb Chops with Greens and Sorrel Salsa Verde ------------------------------------------------------------ Ingredients: ['8 1 1/2-2"-thick lamb loin chops (about 3 pounds)', 'Kosher salt, freshly ground pepper', '2/3 cup sorrel or spinach leaves', '1/3 cup mint leaves', '1/3 cup parsley leaves', '1 tablespoon finely grated lemon zest', '1 teaspoon crushed red pepper flakes', '5 garlic cloves, thinly sliced, divided',... ------------------------------------------------------------ Instructions: Pat lamb dry; season with salt and pepper. Let sit at room temperature 1 hour. Meanwhile, puree sorrel, mint, parsley, lemon zest, red pepper flakes, about 1 garlic clove, and 1/3 cup oil in a blender, scraping down sides as needed, until mixture is smooth. Season salsa verde with salt and pepper; set aside. Prepare grill for high heat (or heat a grill pan over high). Grill chops, turning every 2 ... ================================================================================ RANG #2: 0.6049 Beer-Braised Beef and Onions ------------------------------------------------------------ Ingredients: ['3 pounds onions', '1 (5-pound) boneless beef chuck roast, tied', '2 tablespoons vegetable oil, divided', '2 Turkish bay leaves or 1 California', '2 (12-ounces) bottles pilsner-style beer such as Budweiser', '2 tablespoons red-wine vinegar'] ------------------------------------------------------------ Instructions: Halve onions lengthwise, then slice lengthwise 1/4 inch thick. Pat beef dry and season all over with 2 1/2 teaspoons salt and 1 teaspoon pepper. Heat 1 tablespoon oil in a wide 5-to 6-quart heavy pot over medium-high heat until it shimmers. Brown beef on all sides, about 15 minutes, then transfer to a plate. Cook onions with bay leaves and 1/2 teaspoon salt in remaining tablespoon oil in pot, scra... ================================================================================ RANG #3: 0.5989 Instant-Pot Red Chicken Enchiladas ------------------------------------------------------------ Ingredients: ['1 pound boneless, skinless chicken thighs', '2 cups Ancho Chile Sauce', '1/2 cup chopped onion', '12 corn tortillas', '3 tablespoons vegetable oil or nonstick cooking spray', '1/2 cup crumbled queso fresco', '2 cups shredded Monterey Jack cheese'] ------------------------------------------------------------ Instructions: Put the chicken thighs in the Instant Pot. Pour in the ancho chile sauce. Lock the lid into place. Select Manual; adjust the pressure to High and the time to 10 minutes. After cooking, naturally release the pressure for 10 minutes, then quick release any remaining pressure. Unlock and remove the lid. Transfer the chicken to a bowl and let it cool for a few minutes. Keep the sauce warm. When the ch... ================================================================================ ================================================== TEST IMAGE 2.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.5560 Glazed Cinnamon-Cardamom Buns ------------------------------------------------------------ Ingredients: ['1 cup whole milk', '1 Tbsp. active dry yeast', '1 large egg', '1 large egg yolk', '3 1/2 cups (475 g) all-purpose flour', '1/2 cup (105 g) granulated sugar', '1 1/2 tsp. (3 g) ground cardamom', '1 tsp. kosher salt', '6 Tbsp. room temperature unsalted butter, plus more for bowl', '6 Tbsp. unsalted ... ------------------------------------------------------------ Instructions: Heat milk in a small saucepan over low until just warm; an instant-read thermometer should register 105°F–115°F. Pour into the large bowl of a stand mixer. Whisk in yeast and let sit until foamy, 10–15 minutes. You should see a layer of foam on the surface; this means that the yeast is active. Add egg, egg yolk, flour, granulated sugar, cardamom, and salt to yeast mixture and mix with dough hook o... ================================================================================ RANG #2: 0.5442 Crunchy Chili Onion Rings ------------------------------------------------------------ Ingredients: ['3 cups all purpose flour', '2 tablespoons chili powder', '4 1/2 teaspoons salt', '1 12-ounce bottle dark beer (preferably Mexican)', '12 6-inch corn tortillas, coarsely torn', '2 large onions, peeled', 'Vegetable oil (for deep-frying)'] ------------------------------------------------------------ Instructions: Mix flour, chili powder, and salt in medium bowl. Pour beer into small bowl. Finely grind tortillas in processor; transfer to deep bowl. Line large baking sheet with foil. Cut onions crosswise into 1/2- to 3/4-inch-thick rounds. Separate rounds into rings. Dip 1 onion ring into flour mixture, then beer, flour again and beer again, then add to bowl with ground tortillas and toss to coat. Using wood... ================================================================================ RANG #3: 0.4912 Plain Bagels ------------------------------------------------------------ Ingredients: ['1 envelope active dry yeast (1/4 ounce)', '1 1/2 tablespoons plus 1/4 cup honey, divided', '1 tablespoon non-diasatic malt powder (such as King Arthur brand)', '3 tablespoons neutral oil, plus more for greasing', '9 cups high gluten flour such as bread flour', '2 teaspoons kosher salt'] ------------------------------------------------------------ Instructions: In the bowl of a stand mixer, whisk together 2 2/3 cups water, yeast, and 1 1/2 tablespoons honey. Let stand for 5 minutes. Add malt, oil, flour, and salt and mix with the dough hook attachment at low speed for about 10 minutes, or until dough can "pull a window." To test, pinch off a small ball of dough and pull into a thin, see-through membrane without it tearing. If it tears, mix another minute... ================================================================================ ================================================== TEST IMAGE 3.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.6839 Caramel Apple Cupcakes ------------------------------------------------------------ Ingredients: ['2 1/2 cups all-purpose flour', '3 tsp. baking powder', '1 tsp. ground cinnamon', '1/2 tsp. salt', '16 Tbsp. unsalted butter, at room temperature (European style recommended)', '2 cups sugar', '4 large eggs, at room temperature', '1/3 cup hot water', '2 1/2 cups freshly grated apples (4 to 5 medium... ------------------------------------------------------------ Instructions: Preheat the oven to 350°F. Line a standard cupcake pan with twelve paper baking cups, and a second pan with six baking cups, or grease pans with butter if not using baking cups. Sift together the flour, baking powder, cinnamon, and salt on a sheet of parchment paper or wax paper and set aside. Place the butter in the bowl of a stand mixer or in a bowl with a handheld electric mixer. Beat on medium... ================================================================================ RANG #2: 0.6076 Sweet Potato Cupcakes ------------------------------------------------------------ Ingredients: ['1/2 cup oat flour', '6 tablespoons all-purpose flour', '1/2 teaspoon baking soda', '6 tablespoons granulated sugar', '2 tablespoons unsalted butter, at room temperature', '1 egg white', '1/4 cup cooked or canned sweet potato', '1/4 teaspoon vanilla extract', '2 tablespoons skim milk', "1/4 cup con... ------------------------------------------------------------ Instructions: Heat oven to 350°F. Line 16 cups of two 12-cup mini-muffin pans with mini-muffin papers. In a bowl, combine flours and baking soda. In another bowl, cream together sugar and butter with an electric mixer on medium. Add egg white, sweet potato and vanilla; beat on low until well combined. Add flour mixture and milk; beat on low until just combined. Do not overmix. Fill muffin cups 2/3 full. Bake un... ================================================================================ RANG #3: 0.5712 Double Chocolate Cupcakes With Salted Chia Pudding Frosting ------------------------------------------------------------ Ingredients: ['1 (14-ounce) can full-fat coconut milk', '5 Medjool dates, pitted and chopped', '1/4–1/2 teaspoon of sea salt', '5 tablespoons chia seeds', '1 cup unsweetened almond milk', '1 tablespoon apple cider vinegar', '1 tablespoon flax meal', '3 tablespoons water', '1 cup chickpea (garbanzo bean) flour', ... ------------------------------------------------------------ Instructions: Combine the coconut milk, dates, and salt in a blender and blend until the mixture is smooth. This might take a few minutes. If end up with small bits of dates, that's OK. But you don't want to see chunks or even white slivers of the coconut milk. The color should be a very light tan. Taste, and if you want to add more salt, do it now and then blend for a few more seconds (I really like the saltin... ================================================================================ ================================================== TEST IMAGE 4.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.6819 Fried Egg on Toast with Salted Herb Butter and Radishes ------------------------------------------------------------ Ingredients: ['3/4 cup mixed tender herbs (such as parsley, dill, basil, chives, and/or tarragon), plus sprigs for serving', '1/2 cup (1 stick) unsalted butter, room temperature, cut into pieces', '1/2 teaspoon finely grated lemon zest', '2 teaspoons fresh lemon juice, plus more for drizzling', '1 teaspoon koshe... ------------------------------------------------------------ Instructions: Pulse 3/4 cup herbs in a food processor until finely chopped. Add butter, lemon zest, 2 tsp. lemon juice, 1 tsp. salt, and 1/2 tsp. pepper. Pulse to bring together. Place radishes in a medium bowl; drizzle with a little lemon juice and add a pinch of salt. Toss to combine. To serve, spread toast with herb butter and top each with a fried egg. Scatter dressed radishes over top along with a few spri... ================================================================================ RANG #2: 0.6660 Deep Dish Quiche with Garnishes ------------------------------------------------------------ Ingredients: ['1 full recipe Our Favorite Pie Dough', 'All-purpose flour (for surface)', '2 shallots, finely chopped', '2 Tbsp. extra-virgin olive oil', '1 1/2 tsp. kosher salt, divided', '1 Tbsp. finely chopped chives', '8 large eggs', '1 3/4 cups half-and-half', '1/2 tsp. freshly ground black pepper', '1 cup c... ------------------------------------------------------------ Instructions: Roll dough to a 15" round on a lightly floured work surface; trim edges. Fit dough into a 9" springform pan, letting dough fold over itself slightly around the edges so that it creates a gently pleated, rippled effect (the dough will extend above top of pan). Place pan on a rimmed baking sheet. Chill 1 hour, or freeze 20 minutes. Preheat oven to 350°F. Dock bottom surface of dough with a fork. Lin... ================================================================================ RANG #3: 0.6623 Honey-Mustard Chicken-Sausage Kebabs ------------------------------------------------------------ Ingredients: ['6 tablespoons Dijon mustard', '2 tablespoons honey', '1 tablespoon mayonnaise', '4 fully cooked sun-dried-tomato chicken sausages, each cut into 6 pieces', '24 mini bell peppers', 'Olive oil', 'Metal skewers'] ------------------------------------------------------------ Instructions: Prepare barbecue (medium-high heat). Whisk mustard, honey, and mayonnaise in small bowl to blend. Thread 3 sausage pieces alternatively with 3 peppers onto each of 8 skewers and place on baking sheet. Brush with milk; sprinkle with salt and pepper. Grill skewers until vegetables are lightly charred and crisp-tender and sausage is heated through, turning occasionally and brushing with mustard mixtu... ================================================================================ ================================================== TEST IMAGE 5.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.6865 Farro with Pistachios, Mixed Herbs, and Golden Raisins ------------------------------------------------------------ Ingredients: ['2 cups farro', '1/2 teaspoon kosher salt, plus more', '1/2 cup pistachios', '1 1/2 teaspoons finely grated lemon zest', '3 tablespoons fresh lemon juice', '1 teaspoon finely grated peeled ginger', '1/2 teaspoon sugar', '1/3 cup grapeseed oil or olive oil', 'Freshly ground black pepper', '1 serrano... ------------------------------------------------------------ Instructions: Preheat oven to 350°F. Rinse farro under cold water. Cook in a large pot of boiling salted water, skimming surface occasionally, until tender, 20–25 minutes. Meanwhile, toast pistachios on a rimmed baking sheet, tossing once, until golden brown, 8–10 minutes. Let cool, then coarsely chop. Whisk lemon zest, lemon juice, ginger, sugar, and 1/2 tsp. salt in a medium bowl. Whisking constantly, gradual... ================================================================================ RANG #2: 0.6827 Pasta with Pesto My Way ------------------------------------------------------------ Ingredients: ['2 medium zucchini (about 1/2 pound total)', '1/2 pound boiling potatoes, peeled', '3/4 pound rotini or penne', '6 ounces green beans, trimmed and halved (about 1 1/2 cups)', '2 ears corn, kernels cut off (about 1 1/2 cups)', '1 tablespoon olive oil', '1 pint cherry tomatoes (12 to 14 ounces)', "1 ... ------------------------------------------------------------ Instructions: Cut zucchini into thin rounds with slicer. Fit slicer with julienne attachment and cut potatoes into 1/4-inch-thick matchsticks. Start cooking rotini in a pasta pot of boiling salted water (3 tablespoons salt for 6 quarts water) according to package instructions. Six minutes before pasta is done, stir in potatoes and green beans. Two minutes before pasta is done, stir in corn and zucchini. Meanwhi... ================================================================================ RANG #3: 0.6611 Cavatappi with Broccolini, Brown Butter, and Sage ------------------------------------------------------------ Ingredients: ['Kosher salt', '2 bunches Broccolini (about 1 pound), ends trimmed, split lengthwise into halves or thirds depending on the thickness (or substitute broccoli rabe or broccoli)', '2 tablespoons extra-virgin olive oil', '2 large cloves garlic, minced', '1/4 teaspoon crushed red pepper flakes', 'Fresh... ------------------------------------------------------------ Instructions: Bring a large pot of water to a boil. Fill a large bowl with water and ice and set aside. Add 1 tablespoon kosher salt and the broccolini to the boiling water and cook until crisp-tender, 2 to 3 minutes. Using a spider or slotted spoon, transfer the Broccolini to the ice water to stop the cooking and let cool. Keep the pot of water boiling for the pasta. Drain the broccolini in a colander, cut the... ================================================================================ ================================================== TEST IMAGE 6.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.7219 Spaghetti with Lobster Pomodoro ------------------------------------------------------------ Ingredients: ['Kosher salt', '2 (1 1/4-lb.) live lobsters', '2 Tbsp. extra-virgin olive oil', '2 Tbsp. ghee or unsalted butter', '1/4 small red onion, thinly sliced', '3 garlic cloves, thinly sliced', '2 sprigs basil', '1 (14-oz.) can whole peeled San Marzano tomatoes', '12 oz. spaghetti', '4 oz. nduja, broken i... ------------------------------------------------------------ Instructions: Bring a large pot of salted water to a rolling boil. Working one at a time, cook lobsters 3 minutes, then transfer to a large bowl of ice water. Let cool just until you can comfortably handle them, about 1 minute, then twist off claws where the knuckles meet the body and return them to pot of boiling water. Cook 2 minutes (leave bodies in ice water). Add claws back to ice water and let both claws ... ================================================================================ RANG #2: 0.7216 Rock Shrimp Pasta with Spicy Tomato Sauce ------------------------------------------------------------ Ingredients: ['1 (28-ounce) can whole peeled tomatoes, preferably San Marzano, drained', '1/3 cup olive oil, plus more for drizzling', '1/2 medium fennel bulb, fronds reserved, core removed, bulb thinly sliced', '8 garlic cloves, smashed', '1 Fresno chile, very thinly sliced, divided', '1/4 cup dry white wine', ... ------------------------------------------------------------ Instructions: Set a fine-mesh sieve over a medium bowl. Working over sieve, squeeze tomatoes to release juices and break up flesh. Let tomatoes drain in sieve, collecting juices in bowl, until ready to use. Heat 1/3 cup oil in a large Dutch oven or other heavy pot over medium. Cook fennel, garlic, and half of chile, stirring often, until garlic is golden and fennel is starting to brown around the edges, 5–8 min... ================================================================================ RANG #3: 0.7201 Pasta with Spicy Sausage, Radicchio, and Sun-Dried Tomatoes ------------------------------------------------------------ Ingredients: ['1 pound casarecce or other short pasta', 'Kosher salt', '1/4 cup olive oil', '1 large onion, thinly sliced', '1 pound spicy Italian sausage, casings removed', '1 1/4 pounds radicchio, cored, coarsely chopped (about 7 cups)', '1/2 cup thinly sliced drained oil-packed sun-dried tomatoes', 'Freshly g... ------------------------------------------------------------ Instructions: Cook pasta in a large pot of boiling salted water, stirring occasionally, until al dente. Drain pasta, reserving 3/4 cup pasta cooking liquid; return pasta to pot. Meanwhile, heat oil in a medium skillet over medium-high. Add onion and 1/4 tsp. salt and cook, stirring occasionally, until tender, about 5 minutes (do not brown). Add sausage and cook, stirring occasionally and breaking up sausage int... ================================================================================ ================================================== TEST IMAGE 7.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.5833 Spice-Roasted Porterhouse Steaks ------------------------------------------------------------ Ingredients: ['Two 2-inch-thick porterhouse steaks', 'Olive oil', '1 1/2 tablespoons Coriander-Herb Spice Rub'] ------------------------------------------------------------ Instructions: Place two 2-inch-thick porterhouse steaks on a heavy rimmed baking sheet. Brush both sides with a full-flavored olive oil, then rub 1 1/2 tablespoons of the Coriander-Herb Spice Rub into each side of each steak. Preheat the oven to 450°F. Roast steaks until a meat thermometer inserted into the center registers 125°F for rare, about 25 minutes. Transfer the steaks to a cutting board, cover loosely ... ================================================================================ RANG #2: 0.5763 Pot Roast with Caramelized Onions and Roasted Carrots ------------------------------------------------------------ Ingredients: ['1/2 cup canola oil', 'Kosher salt and freshly ground black pepper', '5 pounds boneless short ribs, denuded (all surface fat removed; have your butcher do this)', '1 cup dry sherry', '4 carrots, peeled and roughly chopped', '2 large onions, peeled and roughly chopped', '8 stalks celery, peeled and ... ------------------------------------------------------------ Instructions: Position racks in upper and lower thirds of oven and preheat to 350°F. Season beef liberally with salt and pepper. In large Dutch oven or heavy ovenproof pot over moderately high heat, heat oil until hot but not smoking. Add beef and sear until dark brown and crisp on both sides, about 10 minutes total. Transfer beef to large plate. Pour off oil in pan and discard. Add sherry and simmer uncovered,... ================================================================================ RANG #3: 0.5723 Wine-Braised Brisket with Tart Cherries ------------------------------------------------------------ Ingredients: ["1/4 cup matzoh cake meal (see Cooks' notes)", 'Kosher or fine salt', 'Freshly ground black pepper', '1 (6- to 6 1/2-pound) first- or second-cut beef brisket', '3 to 4 tablespoons vegetable oil', '16 medium shallots (about 1 pound); peeled, leaving root ends intact', '3 large garlic cloves, finely ... ------------------------------------------------------------ Instructions: Heat oven to 350°F with rack in middle. Whisk together matzoh meal with 1 tablespoon kosher salt (2 teaspoons fine) and 1/2 teaspoon pepper. Pat brisket dry and dredge in matzoh mixture, shaking off excess. Set roasting pan across 2 burners and in it heat 3 tablespoons oil over medium-high heat until oil shimmers. Brown brisket (fat side down first if using first cut) on both sides, 3 to 5 minutes... ================================================================================ ================================================== TEST IMAGE 8.jpg ================================================== Recherche de 3 recettes similaires... 3 recipes similaires trouvées
================================================================================ RECETTES SIMILAIRES ================================================================================ RANG #1: 0.5943 Wine-Braised Brisket with Tart Cherries ------------------------------------------------------------ Ingredients: ["1/4 cup matzoh cake meal (see Cooks' notes)", 'Kosher or fine salt', 'Freshly ground black pepper', '1 (6- to 6 1/2-pound) first- or second-cut beef brisket', '3 to 4 tablespoons vegetable oil', '16 medium shallots (about 1 pound); peeled, leaving root ends intact', '3 large garlic cloves, finely ... ------------------------------------------------------------ Instructions: Heat oven to 350°F with rack in middle. Whisk together matzoh meal with 1 tablespoon kosher salt (2 teaspoons fine) and 1/2 teaspoon pepper. Pat brisket dry and dredge in matzoh mixture, shaking off excess. Set roasting pan across 2 burners and in it heat 3 tablespoons oil over medium-high heat until oil shimmers. Brown brisket (fat side down first if using first cut) on both sides, 3 to 5 minutes... ================================================================================ RANG #2: 0.5848 Yum Yum Nut Sweets ------------------------------------------------------------ Ingredients: ['1/2 teaspoon safflower oil', '1 cup whole roasted unsalted almonds', '1 cup whole walnuts', '1/3 cup sugar', '2 egg whites'] ------------------------------------------------------------ Instructions: Preheat the oven to 300°F. Grease a 7- or 8-inch square baking pan with the oil. Set aside. Place the nuts in a food processor and chop until the mixture resembles coarse sand. Put the contents into a small mixing bowl. Stir in the sugar and egg whites. With a spatula, scrape the contents of the bowl into the oiled pan and press into an even layer. Bake for 1 hour for soft cookies or 1 1/2 hours f... ================================================================================ RANG #3: 0.5762 Banana Bread ------------------------------------------------------------ Ingredients: ['1 stick butter, melted, plus softened butter for greasing', '2 cups flour', '1/2 teaspoon salt', '1 1/2 teaspoons baking powder', '1 cup sugar', '3 very ripe bananas, mashed with a fork until smooth', '2 eggs', '1 teaspoon vanilla extract', '1/2 cup chopped walnuts (optional)', '1/2 cup shredded u... ------------------------------------------------------------ Instructions: Heat the oven to 350°F. Grease a 9x5-inch loaf pan with softened butter. Whisk together the flour, salt, baking powder, and sugar in a large bowl. Mix together the melted butter and mashed bananas in a separate bowl. Beat in the eggs and vanilla until well combined. Stir this mixture into the dry ingredients just enough to combine everything. Gently fold in the nuts and coconut if you’re using the... ================================================================================